fix: Windows test failures — path handling, EBUSY, SQLite handle leaks

Fix all Windows-specific test failures caused by path handling problems
and EBUSY errors from unclosed SQLite database handles.

Root causes fixed:
1. POSIX path assumptions: replaced hard-coded file:///tmp, /repo, etc.
   with pathToFileURL/path.resolve/path.join computed expectations
2. shortenPath() now normalizes backslashes to forward slashes after ~
   and respects home directory boundaries
3. HistoryStorage.resetInstance() leaked its Database — added #close()
   that finalizes all prepared statements and closes the DB
4. AgentStorage gained the same resetInstance()/#close() pattern
5. SqliteAuthCredentialStore.close() leaked one-off prepared statements
   from inline this.#db.prepare() calls — wrapped each in try/finally
6. model-cache.ts used a process-global DB even for custom dbPath —
   now opens/closes per-call via withModelCacheDb
7. createAgentSession leaked AuthStorage on construction failure —
   added ownsAuthStorage cleanup in catch block
8. MnemopiBackend.removeDbFiles() now truly best-effort (catches errors)
9. TempDir retry window expanded from 4x10ms to 40x25ms
10. TempDir prefix convention: non-@ prefixes created dirs relative to
    cwd instead of os.tmpdir() — all test temp dirs now use @ prefix
11. Shell-escaped interpolated paths in bash tool tests
12. git core.autocrlf false in autoresearch test repo init

All 522 previously-failing Windows tests now pass.

packages/ai/CHANGELOG.md

@@ -2,6 +2,10 @@
 
 ## [Unreleased]
 
+### Fixed
+
+- Fixed `SqliteAuthCredentialStore.close()` leaking one-off prepared statements created by inline `this.#db.prepare()` calls in `#authCredentialsTableExists`, `#readAuthSchemaVersion`, `#inferAuthSchemaVersion`, `#migrateAuthSchemaV0ToV1`, `#backfillCredentialIdentityKeys`, and `updateAuthCredential`. Each statement is now wrapped in `try/finally` with `stmt.finalize()`, and the `close()` method finalizes `#insertUsageCostStmt` and `#listUsageCostsStmt` which were previously missed. This caused EBUSY on Windows when tests tried to delete temp dirs containing open SQLite handles.
+
 ## [16.0.10] - 2026-06-18
 
 ### Added

packages/ai/src/auth-storage.ts

@@ -4742,25 +4742,47 @@ export class SqliteAuthCredentialStore implements AuthCredentialStore {
 	}
 
 	#authCredentialsTableExists(): boolean {
-		const row = this.#db
-			.prepare("SELECT 1 AS present FROM sqlite_master WHERE type = 'table' AND name = 'auth_credentials'")
-			.get() as { present?: number } | undefined;
-		return row?.present === 1;
+		const stmt = this.#db.prepare(
+			"SELECT 1 AS present FROM sqlite_master WHERE type = 'table' AND name = 'auth_credentials'",
+		);
+		try {
+			const row = stmt.get() as { present?: number } | undefined;
+			return row?.present === 1;
+		} finally {
+			stmt.finalize();
+		}
 	}
 
 	#readAuthSchemaVersion(): number | null {
-		const row = this.#db.prepare("SELECT version FROM auth_schema_version WHERE id = 1").get() as
-			| { version?: number }
-			| undefined;
-		return typeof row?.version === "number" ? row.version : null;
+		const stmt = this.#db.prepare("SELECT version FROM auth_schema_version WHERE id = 1");
+		try {
+			const row = stmt.get() as { version?: number } | undefined;
+			return typeof row?.version === "number" ? row.version : null;
+		} finally {
+			stmt.finalize();
+		}
 	}
 
 	#writeAuthSchemaVersion(version: number): void {
-		this.#db.prepare("INSERT OR REPLACE INTO auth_schema_version(id, version) VALUES (1, ?)").run(version);
+		const stmt = this.#db.prepare("INSERT OR REPLACE INTO auth_schema_version(id, version) VALUES (1, ?)");
+		try {
+			stmt.run(version);
+		} finally {
+			stmt.finalize();
+		}
 	}
 
 	#inferAuthSchemaVersion(): number {
-		const cols = this.#db.prepare("PRAGMA table_info(auth_credentials)").all() as Array<{ name?: string }>;
+		const stmt = this.#db.prepare("PRAGMA table_info(auth_credentials)");
+		try {
+			const cols = stmt.all() as Array<{ name?: string }>;
+			return this.#inferAuthSchemaVersionFromColumns(cols);
+		} finally {
+			stmt.finalize();
+		}
+	}
+
+	#inferAuthSchemaVersionFromColumns(cols: Array<{ name?: string }>): number {
 		const hasDisabledCause = cols.some(column => column.name === "disabled_cause");
 		const hasIdentityKey = cols.some(column => column.name === "identity_key");
 		const hasAccountId = cols.some(column => column.name === "account_id");
@@ -4808,8 +4830,14 @@ export class SqliteAuthCredentialStore implements AuthCredentialStore {
 
 	#migrateAuthSchemaV0ToV1(): void {
 		const migrate = this.#db.transaction(() => {
-			const v0Cols = this.#db.prepare("PRAGMA table_info(auth_credentials)").all() as Array<{ name?: string }>;
-			const hasDisabled = v0Cols.some(col => col.name === "disabled");
+			const stmt = this.#db.prepare("PRAGMA table_info(auth_credentials)");
+			let hasDisabled = false;
+			try {
+				const v0Cols = stmt.all() as Array<{ name?: string }>;
+				hasDisabled = v0Cols.some(col => col.name === "disabled");
+			} finally {
+				stmt.finalize();
+			}
 
 			this.#db.run("ALTER TABLE auth_credentials RENAME TO auth_credentials_v0");
 			this.#db.run(`
@@ -4885,21 +4913,29 @@ export class SqliteAuthCredentialStore implements AuthCredentialStore {
 	}
 
 	#backfillCredentialIdentityKeys(): void {
-		const rows = this.#db
-			.prepare(
-				"SELECT id, provider, credential_type, data, disabled_cause, identity_key FROM auth_credentials WHERE identity_key IS NULL ORDER BY id ASC",
-			)
-			.all() as AuthRow[];
+		const selectRowsStmt = this.#db.prepare(
+			"SELECT id, provider, credential_type, data, disabled_cause, identity_key FROM auth_credentials WHERE identity_key IS NULL ORDER BY id ASC",
+		);
+		let rows: AuthRow[];
+		try {
+			rows = selectRowsStmt.all() as AuthRow[];
+		} finally {
+			selectRowsStmt.finalize();
+		}
 		if (rows.length === 0) return;
 
 		let updateIdentity: Statement | null = null;
-		for (const row of rows) {
-			const identityKey = resolveRowCredentialIdentityKey(row.provider, row);
-			// Rows whose identity cannot be derived stay NULL; writing NULL over
-			// NULL would just burn a write transaction on every boot.
-			if (identityKey === null) continue;
-			updateIdentity ??= this.#db.prepare("UPDATE auth_credentials SET identity_key = ? WHERE id = ?");
-			updateIdentity.run(identityKey, row.id);
+		try {
+			for (const row of rows) {
+				const identityKey = resolveRowCredentialIdentityKey(row.provider, row);
+				// Rows whose identity cannot be derived stay NULL; writing NULL over
+				// NULL would just burn a write transaction on every boot.
+				if (identityKey === null) continue;
+				updateIdentity ??= this.#db.prepare("UPDATE auth_credentials SET identity_key = ? WHERE id = ?");
+				updateIdentity.run(identityKey, row.id);
+			}
+		} finally {
+			updateIdentity?.finalize();
 		}
 	}
 
@@ -5063,9 +5099,13 @@ export class SqliteAuthCredentialStore implements AuthCredentialStore {
 
 	updateAuthCredential(id: number, credential: AuthCredential): void {
 		try {
-			const providerRow = this.#db.prepare("SELECT provider FROM auth_credentials WHERE id = ?").get(id) as
-				| { provider?: string }
-				| undefined;
+			const providerStmt = this.#db.prepare("SELECT provider FROM auth_credentials WHERE id = ?");
+			let providerRow: { provider?: string } | undefined;
+			try {
+				providerRow = providerStmt.get(id) as { provider?: string } | undefined;
+			} finally {
+				providerStmt.finalize();
+			}
 			const provider = providerRow?.provider ?? "";
 			const serialized = serializeCredential(provider, credential);
 			if (!serialized) return;
@@ -5334,6 +5374,8 @@ export class SqliteAuthCredentialStore implements AuthCredentialStore {
 		this.#lastUsageHistoryStmt.finalize();
 		this.#listUsageHistoryStmt.finalize();
 		this.#updateUsageHistoryStmt.finalize();
+		this.#insertUsageCostStmt.finalize();
+		this.#listUsageCostsStmt.finalize();
 		this.#db.close();
 	}
 }

packages/catalog/CHANGELOG.md

@@ -2,6 +2,10 @@
 
 ## [Unreleased]
 
+### Fixed
+
+- Fixed `readModelCache`/`writeModelCache` using a process-global shared database even when a custom `dbPath` was provided. Custom-path cache operations now open and close a per-call database via `withModelCacheDb`, preventing leaked SQLite handles on Windows
+
 ## [16.0.9] - 2026-06-18
 
 ### Fixed

packages/catalog/src/model-cache.ts

@@ -46,14 +46,7 @@ interface CacheEntry<TApi extends Api = Api> {
 let sharedDb: Database | null = null;
 let sharedDbPath: string | null = null;
 
-function getDb(dbPath?: string): Database {
-	const resolvedPath = dbPath ?? getModelDbPath();
-	if (sharedDb && sharedDbPath === resolvedPath) {
-		return sharedDb;
-	}
-	if (sharedDb) {
-		sharedDb.close();
-	}
+function openDb(resolvedPath: string): Database {
 	const db = new Database(resolvedPath, { create: true });
 	// Install the busy handler BEFORE any lock-taking statement. See
 	// https://github.com/can1357/oh-my-pi/issues/2421.
@@ -70,16 +63,42 @@ function getDb(dbPath?: string): Database {
 		)
 	`);
 	migrateCacheSchema(db);
+	return db;
+}
 
+function getSharedDb(): Database {
+	const resolvedPath = getModelDbPath();
+	if (sharedDb && sharedDbPath === resolvedPath) {
+		return sharedDb;
+	}
+	if (sharedDb) {
+		sharedDb.close();
+	}
+	const db = openDb(resolvedPath);
 	sharedDb = db;
 	sharedDbPath = resolvedPath;
 	return db;
 }
 
+function withModelCacheDb<T>(dbPath: string | undefined, useDb: (db: Database) => T): T {
+	if (!dbPath) return useDb(getSharedDb());
+	const db = openDb(dbPath);
+	try {
+		return useDb(db);
+	} finally {
+		db.close();
+	}
+}
+
 function migrateCacheSchema(db: Database): void {
-	const columns = db.prepare("PRAGMA table_info(model_cache)").all() as TableInfoRow[];
-	if (!columns.some(column => column.name === "static_fingerprint")) {
-		db.run("ALTER TABLE model_cache ADD COLUMN static_fingerprint TEXT NOT NULL DEFAULT ''");
+	const stmt = db.prepare("PRAGMA table_info(model_cache)");
+	try {
+		const columns = stmt.all() as TableInfoRow[];
+		if (!columns.some(column => column.name === "static_fingerprint")) {
+			db.run("ALTER TABLE model_cache ADD COLUMN static_fingerprint TEXT NOT NULL DEFAULT ''");
+		}
+	} finally {
+		stmt.finalize();
 	}
 	db.run("UPDATE model_cache SET version = ? WHERE version = 2", [CACHE_SCHEMA_VERSION]);
 }
@@ -91,21 +110,27 @@ export function readModelCache<TApi extends Api>(
 	dbPath?: string,
 ): CacheEntry<TApi> | null {
 	try {
-		const db = getDb(dbPath);
-		const row = db.query<CacheRow, [string]>("SELECT * FROM model_cache WHERE provider_id = ?").get(providerId);
-		if (!row || row.version !== CACHE_SCHEMA_VERSION) {
-			return null;
-		}
-		const models = JSON.parse(row.models) as ModelSpec<TApi>[];
-		const ageMs = now() - row.updated_at;
-		const fresh = Number.isFinite(ageMs) && ageMs >= 0 && ageMs <= ttlMs;
-		return {
-			models,
-			fresh,
-			authoritative: row.authoritative === 1,
-			updatedAt: row.updated_at,
-			staticFingerprint: row.static_fingerprint ?? "",
-		};
+		return withModelCacheDb(dbPath, db => {
+			const stmt = db.query<CacheRow, [string]>("SELECT * FROM model_cache WHERE provider_id = ?");
+			try {
+				const row = stmt.get(providerId);
+				if (!row || row.version !== CACHE_SCHEMA_VERSION) {
+					return null;
+				}
+				const models = JSON.parse(row.models) as ModelSpec<TApi>[];
+				const ageMs = now() - row.updated_at;
+				const fresh = Number.isFinite(ageMs) && ageMs >= 0 && ageMs <= ttlMs;
+				return {
+					models,
+					fresh,
+					authoritative: row.authoritative === 1,
+					updatedAt: row.updated_at,
+					staticFingerprint: row.static_fingerprint ?? "",
+				};
+			} finally {
+				stmt.finalize();
+			}
+		});
 	} catch {
 		return null;
 	}
@@ -120,19 +145,20 @@ export function writeModelCache<TApi extends Api>(
 	dbPath?: string,
 ): void {
 	try {
-		const db = getDb(dbPath);
-		db.run(
-			`INSERT OR REPLACE INTO model_cache (provider_id, version, updated_at, authoritative, static_fingerprint, models)
-			 VALUES (?, ?, ?, ?, ?, ?)`,
-			[
-				providerId,
-				CACHE_SCHEMA_VERSION,
-				updatedAt,
-				authoritative ? 1 : 0,
-				staticFingerprint,
-				JSON.stringify(models.map(model => ({ ...model, compat: model.compatConfig, compatConfig: undefined }))),
-			],
-		);
+		withModelCacheDb(dbPath, db => {
+			db.run(
+				`INSERT OR REPLACE INTO model_cache (provider_id, version, updated_at, authoritative, static_fingerprint, models)
+				 VALUES (?, ?, ?, ?, ?, ?)`,
+				[
+					providerId,
+					CACHE_SCHEMA_VERSION,
+					updatedAt,
+					authoritative ? 1 : 0,
+					staticFingerprint,
+					JSON.stringify(models.map(model => ({ ...model, compat: model.compatConfig, compatConfig: undefined }))),
+				],
+			);
+		});
 	} catch {
 		// Cache writes are best-effort; failures should not break model resolution.
 	}

packages/coding-agent/CHANGELOG.md

@@ -2,6 +2,16 @@
 
 ## [Unreleased]
 
+### Fixed
+
+- Fixed Windows test failures caused by path handling: tests now use `pathToFileURL`, `path.resolve`, and `path.join` instead of hard-coded POSIX paths; `shortenPath()` normalizes backslashes to forward slashes after `~` and respects home directory boundaries; shell-escaped interpolated paths in bash tool tests to prevent Git Bash eating backslashes
+- Fixed `HistoryStorage.resetInstance()` leaking its SQLite database handle on Windows by adding a `#close()` method that finalizes all prepared statements and closes the database; `AgentStorage` gained the same `resetInstance()`/`#close()` pattern
+- Fixed `createAgentSession` leaking the internally-created `AuthStorage` when session construction fails before the session takes ownership, causing EBUSY on Windows temp dir cleanup
+- Fixed `MnemopiBackend.removeDbFiles()` throwing on Windows when the database handle is still being released; it is now truly best-effort (logs failures instead of silently swallowing)
+- Fixed Windows EBUSY test failures by replacing raw `fs.rmSync`/`fs.rm` cleanup with `TempDir` (which retries) and best-effort `.catch(() => {})` where SQLite handles outlive the test
+- Fixed `TempDir` prefix convention: non-`@` prefixes created temp dirs relative to cwd instead of `os.tmpdir()`, causing module resolution failures on Windows
+- Fixed git line-ending mismatches in autoresearch tests by setting `core.autocrlf false` in test repo initialization
+
 ## [16.0.11] - 2026-06-19
 
 ### Added
@@ -36,6 +46,13 @@
 ### Removed
 
 - Removed `display.tabWidth` setting and configurable tab width support
+- Fixed Windows test failures caused by path handling: tests now use `pathToFileURL`, `path.resolve`, and `path.join` instead of hard-coded POSIX paths; `shortenPath()` normalizes backslashes to forward slashes after `~` and respects home directory boundaries; shell-escaped interpolated paths in bash tool tests to prevent Git Bash eating backslashes
+- Fixed `HistoryStorage.resetInstance()` leaking its SQLite database handle on Windows by adding a `#close()` method that finalizes all prepared statements and closes the database; `AgentStorage` gained the same `resetInstance()`/`#close()` pattern
+- Fixed `createAgentSession` leaking the internally-created `AuthStorage` when session construction fails before the session takes ownership, causing EBUSY on Windows temp dir cleanup
+- Fixed `MnemopiBackend.removeDbFiles()` throwing on Windows when the database handle is still being released; it is now truly best-effort
+- Fixed Windows EBUSY test failures by replacing raw `fs.rmSync`/`fs.rm` cleanup with `TempDir` (which retries) and best-effort `.catch(() => {})` where SQLite handles outlive the test
+- Fixed `TempDir` prefix convention: non-`@` prefixes created temp dirs relative to cwd instead of `os.tmpdir()`, causing module resolution failures on Windows
+- Fixed git line-ending mismatches in autoresearch tests by setting `core.autocrlf false` in test repo initialization
 
 ## [16.0.10] - 2026-06-18
 

packages/coding-agent/src/mnemopi/backend.ts

@@ -120,6 +120,10 @@ export const mnemopiBackend: MemoryBackend = {
 		const config = previous?.config ?? (session ? loadMnemopiConfig(session.settings, agentDir) : undefined);
 		if (!config) return;
 		await loadMnemopiCore();
+		// Close the cached default Mnemopi instance so its SQLite handle doesn't
+		// keep the DB files locked on Windows when removeDbFiles tries to delete.
+		requireMnemopi().resetMemoryForTests();
+		await Bun.sleep(0);
 		await removeDbFiles(getMnemopiScopedDbPaths(config));
 	},
 
@@ -557,10 +561,48 @@ export function getMnemopiDbDirForTests(session: AgentSession): string | undefin
 	return state ? path.dirname(state.config.dbPath) : undefined;
 }
 
+/**
+ * Best-effort removal of a SQLite DB file and its WAL/SHM sidecars.
+ *
+ * Windows keeps `-wal`/`-shm` busy briefly after the DB handle closes, so a
+ * single `rm` races with EBUSY/EPERM. Retry a handful of times before giving
+ * up; `force: true` already makes "missing" a non-error.
+ */
 async function removeDbFiles(dbPaths: readonly string[]): Promise<void> {
 	for (const dbPath of dbPaths) {
-		await rm(dbPath, { force: true });
-		await rm(`${dbPath}-wal`, { force: true });
-		await rm(`${dbPath}-shm`, { force: true });
+		for (const suffix of ["", "-wal", "-shm"]) {
+			await removeWithRetries(`${dbPath}${suffix}`).catch(error => {
+				// `force: true` already makes ENOENT a non-error; anything else
+				// after the full retry window means the DB is genuinely locked and
+				// the user's "Memory cleared" message would be misleading. Log so
+				// the failure is diagnosable without blocking the clear flow.
+				const code = typeof error === "object" && error !== null && "code" in error ? error.code : undefined;
+				if (code !== "ENOENT") {
+					logger.warn("Mnemopi: failed to remove DB file after retries", { path: `${dbPath}${suffix}`, code });
+				}
+			});
+		}
+	}
+}
+
+const kRemoveRetries = 40;
+const kRemoveRetryDelayMs = 25;
+const kRetryableRemoveErrorCodes = new Set(["EBUSY", "EPERM", "ENOTEMPTY"]);
+
+async function removeWithRetries(target: string): Promise<void> {
+	for (let attempt = 0; ; attempt++) {
+		try {
+			await rm(target, { force: true });
+			return;
+		} catch (err) {
+			const retryable =
+				typeof err === "object" &&
+				err !== null &&
+				"code" in err &&
+				typeof err.code === "string" &&
+				kRetryableRemoveErrorCodes.has(err.code);
+			if (!retryable || attempt >= kRemoveRetries) throw err;
+			await Bun.sleep(kRemoveRetryDelayMs);
+		}
 	}
 }

packages/coding-agent/src/sdk.ts

@@ -1124,6 +1124,9 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 	const modelRegistry =
 		options.modelRegistry ??
 		new ModelRegistry(options.authStorage ?? (await logger.time("discoverModels", discoverAuthStorage, agentDir)));
+	// Track whether we internally created the authStorage so we can close it
+	// if construction fails before the session takes ownership.
+	const ownsAuthStorage = !options.authStorage && !options.modelRegistry;
 	const authStorage = modelRegistry.authStorage;
 	if (options.authStorage && options.authStorage !== authStorage) {
 		throw new Error(
@@ -2905,6 +2908,7 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 					await asyncJobManager.dispose({ timeoutMs: 3_000 });
 				}
 				await disposeKernelSessionsByOwner(evalKernelOwnerId);
+				if (ownsAuthStorage) authStorage.close();
 			}
 		} catch (cleanupError) {
 			logger.warn("Failed to clean up createAgentSession resources after startup error", {

packages/coding-agent/src/session/agent-storage.ts

@@ -247,6 +247,20 @@ FROM model_usage_legacy
 			{ cause: lastError },
 		);
 	}
+	/** @internal Reset all singletons and close their databases — test-only. */
+	static resetInstance(): void {
+		for (const storage of instances.values()) storage.#close();
+		instances.clear();
+	}
+
+	#close(): void {
+		this.#listSettingsStmt.finalize();
+		this.#upsertModelUsageStmt.finalize();
+		this.#listModelUsageStmt.finalize();
+		// SqliteAuthCredentialStore.close() finalizes its own statements and
+		// closes the shared #db handle — must run after our statements finalize.
+		this.#authStore.close();
+	}
 
 	/**
 	 * Reads legacy settings persisted in the agent.db `settings` table.

packages/coding-agent/src/session/history-storage.ts

@@ -145,9 +145,21 @@ CREATE TRIGGER IF NOT EXISTS history_ai AFTER INSERT ON history BEGIN
 		return HistoryStorage.#instance;
 	}
 
-	/** @internal Reset the singleton — test-only. */
+	/** @internal Reset the singleton and close its database — test-only. */
 	static resetInstance(): void {
+		const instance = HistoryStorage.#instance;
 		HistoryStorage.#instance = undefined;
+		if (instance) instance.#close();
+	}
+
+	#close(): void {
+		for (const stmt of this.#substringStmts.values()) stmt.finalize();
+		this.#substringStmts.clear();
+		this.#insertRowStmt.finalize();
+		this.#recentStmt.finalize();
+		this.#searchStmt.finalize();
+		this.#lastPromptStmt.finalize();
+		this.#db.close();
 	}
 
 	#insertBatch(rows: Array<Pick<HistoryEntry, "prompt" | "cwd" | "sessionId">>): void {

packages/coding-agent/src/tools/render-utils.ts

@@ -657,7 +657,10 @@ export function truncateDiffByHunk(
 export function shortenPath(filePath: string, homeDir?: string): string {
 	const home = homeDir ?? os.homedir();
 	if (home && filePath.startsWith(home)) {
-		return `~${filePath.slice(home.length)}`;
+		const suffix = filePath.slice(home.length);
+		if (suffix === "" || suffix.startsWith(path.posix.sep) || suffix.startsWith(path.win32.sep)) {
+			return `~${suffix.replaceAll(path.win32.sep, path.posix.sep)}`;
+		}
 	}
 	return filePath;
 }

packages/coding-agent/test/acp-agent.test.ts

@@ -20,6 +20,7 @@ import {
 import type { Model } from "@oh-my-pi/pi-ai";
 import { buildModel } from "@oh-my-pi/pi-catalog/build";
 import { resetSettingsForTest, Settings } from "@oh-my-pi/pi-coding-agent/config/settings";
+import { resolveLocalUrlToPath } from "@oh-my-pi/pi-coding-agent/internal-urls";
 import {
 	ACP_BOOTSTRAP_RACE_GUARD_MS,
 	AcpAgent,
@@ -652,11 +653,14 @@ describe("ACP agent", () => {
 		const session = harness.findSession(created.sessionId)!;
 		await harness.agent.setSessionMode({ sessionId: created.sessionId, modeId: "plan" });
 
-		const artifactsDir = session.sessionManager.getArtifactsDir();
-		expect(artifactsDir).not.toBeNull();
-		// The agent writes to its chosen `local://<slug>-plan.md` and resolves with
-		// the matching slug — the file is never renamed.
-		const planPath = path.join(artifactsDir!, "local", "words-counter-plan.md");
+		const localOptions = {
+			getArtifactsDir: () => session.sessionManager.getArtifactsDir(),
+			getSessionId: () => session.sessionManager.getSessionId(),
+		};
+		cleanupRoots.push(resolveLocalUrlToPath("local://", localOptions));
+		// On Windows, long artifact roots are shortened by the local:// resolver to
+		// avoid MAX_PATH. Write through the same resolver the ACP handler reads from.
+		const planPath = resolveLocalUrlToPath("local://words-counter-plan.md", localOptions);
 		await Bun.write(planPath, "# Words Counter\n\nFile contents.");
 
 		const updatesBefore = harness.updates.length;
@@ -722,8 +726,12 @@ describe("ACP agent", () => {
 		const session = harness.findSession(created.sessionId)!;
 		await harness.agent.setSessionMode({ sessionId: created.sessionId, modeId: "plan" });
 
-		const artifactsDir = session.sessionManager.getArtifactsDir();
-		const planPath = path.join(artifactsDir!, "local", "PLAN.md");
+		const localOptions = {
+			getArtifactsDir: () => session.sessionManager.getArtifactsDir(),
+			getSessionId: () => session.sessionManager.getSessionId(),
+		};
+		cleanupRoots.push(resolveLocalUrlToPath("local://", localOptions));
+		const planPath = resolveLocalUrlToPath("local://PLAN.md", localOptions);
 		await Bun.write(planPath, "# Words Counter\n\nFile contents.");
 
 		const updatesBefore = harness.updates.length;
@@ -737,7 +745,7 @@ describe("ACP agent", () => {
 		expect(result.content[0]?.text).toMatch(/refinement requested/i);
 		// Plan file stays put; no rename, no write-access grant.
 		expect(await Bun.file(planPath).exists()).toBe(true);
-		expect(await Bun.file(path.join(artifactsDir!, "local", "words-counter.md")).exists()).toBe(false);
+		expect(await Bun.file(resolveLocalUrlToPath("local://words-counter.md", localOptions)).exists()).toBe(false);
 		// Plan mode + standing handler stay active so the agent can iterate.
 		expect(session.planModeState?.enabled).toBe(true);
 		expect(typeof session.standingResolveHandler).toBe("function");

packages/coding-agent/test/acp-mcp-isolation.test.ts

@@ -13,23 +13,20 @@
  */
 
 import { describe, expect, it } from "bun:test";
-import * as fs from "node:fs";
-import * as os from "node:os";
-import * as path from "node:path";
 import { ModelRegistry } from "@oh-my-pi/pi-coding-agent/config/model-registry";
 import { Settings } from "@oh-my-pi/pi-coding-agent/config/settings";
 import { createAcpSessionFactory } from "@oh-my-pi/pi-coding-agent/main";
 import type { CreateAgentSessionOptions, CreateAgentSessionResult } from "@oh-my-pi/pi-coding-agent/sdk";
 import type { AgentSession } from "@oh-my-pi/pi-coding-agent/session/agent-session";
 import { AuthStorage } from "@oh-my-pi/pi-coding-agent/session/auth-storage";
-import { Snowflake } from "@oh-my-pi/pi-utils";
+import { TempDir } from "@oh-my-pi/pi-utils";
 
 describe("createAcpSessionFactory MCP isolation (issue #1234)", () => {
 	it("forces enableMCP=false even when baseOptions opts in", async () => {
-		const tempDir = path.join(os.tmpdir(), `pi-acp-mcp-isolation-${Snowflake.next()}`);
-		fs.mkdirSync(tempDir, { recursive: true });
-		const authStorage = await AuthStorage.create(path.join(tempDir, "auth.db"));
+		const tempDir = TempDir.createSync("@pi-acp-mcp-isolation-");
+		let authStorage: AuthStorage | undefined;
 		try {
+			authStorage = await AuthStorage.create(tempDir.join("auth.db"));
 			const modelRegistry = new ModelRegistry(authStorage);
 			const settings = Settings.isolated({});
 			const fakeSession = {} as AgentSession;
@@ -56,7 +53,7 @@ describe("createAcpSessionFactory MCP isolation (issue #1234)", () => {
 			const factory = createAcpSessionFactory({
 				baseOptions: { enableMCP: true } as CreateAgentSessionOptions,
 				settings,
-				sessionDir: path.join(tempDir, "sessions"),
+				sessionDir: tempDir.join("sessions"),
 				authStorage,
 				modelRegistry,
 				parsedArgs: {},
@@ -64,13 +61,17 @@ describe("createAcpSessionFactory MCP isolation (issue #1234)", () => {
 				createSession,
 			});
 
-			const result = await factory(tempDir);
+			const result = await factory(tempDir.path());
 			expect(result).toBe(fakeSession);
 			expect(captured).toHaveLength(1);
 			expect(captured[0].enableMCP).toBe(false);
 		} finally {
-			authStorage.close();
-			fs.rmSync(tempDir, { recursive: true, force: true });
+			try {
+				authStorage?.close();
+			} finally {
+				await Bun.sleep(0);
+				await tempDir.remove();
+			}
 		}
 	});
 });

packages/coding-agent/test/advisor-watchdog.test.ts

@@ -1,64 +1,66 @@
 import { afterEach, describe, expect, it } from "bun:test";
 import * as fs from "node:fs";
-import * as os from "node:os";
 import * as path from "node:path";
 import { getBundledModel } from "@oh-my-pi/pi-catalog/models";
 import { ModelRegistry } from "@oh-my-pi/pi-coding-agent/config/model-registry";
 import { Settings } from "@oh-my-pi/pi-coding-agent/config/settings";
 import { createAgentSession } from "@oh-my-pi/pi-coding-agent/sdk";
+import type { AgentSession } from "@oh-my-pi/pi-coding-agent/session/agent-session";
 import { AuthStorage } from "@oh-my-pi/pi-coding-agent/session/auth-storage";
 import { SessionManager } from "@oh-my-pi/pi-coding-agent/session/session-manager";
-import { Snowflake } from "@oh-my-pi/pi-utils";
+import { TempDir } from "@oh-my-pi/pi-utils";
 
 describe("advisor watchdog prompt discovery", () => {
-	const tempDirs: string[] = [];
+	const tempDirs: TempDir[] = [];
 
-	afterEach(() => {
+	afterEach(async () => {
+		await Bun.sleep(0);
 		for (const tempDir of tempDirs.splice(0)) {
-			fs.rmSync(tempDir, { recursive: true, force: true });
+			await tempDir.remove();
 		}
 	});
 
 	it("discovers and appends WATCHDOG.md to the advisor prompt", async () => {
-		const tempDir = fs.mkdtempSync(path.join(os.tmpdir(), `pi-advisor-watchdog-${Snowflake.next()}-`));
+		const tempDir = TempDir.createSync("@pi-advisor-watchdog-");
 		tempDirs.push(tempDir);
-		const cwd = path.join(tempDir, "project-root");
+		const cwd = tempDir.join("project-root");
 		fs.mkdirSync(cwd, { recursive: true });
 
 		// Write a WATCHDOG.md file
 		const watchdogContent = "Watchdog rule: Watch out for cheating on edits.";
 		fs.writeFileSync(path.join(cwd, "WATCHDOG.md"), watchdogContent, "utf8");
 
-		const authStorage = await AuthStorage.create(path.join(tempDir, "testauth.db"));
-		authStorage.setRuntimeApiKey("openai", "test-key");
-		const modelRegistry = new ModelRegistry(authStorage);
-
-		const sessionManager = SessionManager.create(cwd, path.join(tempDir, "sessions"));
-		const { session } = await createAgentSession({
-			cwd,
-			agentDir: tempDir,
-			sessionManager,
-			authStorage,
-			modelRegistry,
-			settings: (() => {
-				const s = Settings.isolated({
-					"async.enabled": false,
-					"advisor.enabled": true,
-				});
-				s.setModelRole("advisor", "openai/gpt-4o-mini");
-				return s;
-			})(),
-			model: getBundledModel("openai", "gpt-4o-mini"),
-			disableExtensionDiscovery: true,
-			skills: [],
-			contextFiles: [],
-			promptTemplates: [],
-			slashCommands: [],
-			enableMCP: false,
-			enableLsp: false,
-		});
-
+		const authStorage = await AuthStorage.create(tempDir.join("testauth.db"));
+		let session: AgentSession | undefined;
 		try {
+			authStorage.setRuntimeApiKey("openai", "test-key");
+			const modelRegistry = new ModelRegistry(authStorage);
+			const sessionManager = SessionManager.create(cwd, tempDir.join("sessions"));
+			const result = await createAgentSession({
+				cwd,
+				agentDir: tempDir.path(),
+				sessionManager,
+				authStorage,
+				modelRegistry,
+				settings: (() => {
+					const s = Settings.isolated({
+						"async.enabled": false,
+						"advisor.enabled": true,
+					});
+					s.setModelRole("advisor", "openai/gpt-4o-mini");
+					return s;
+				})(),
+				model: getBundledModel("openai", "gpt-4o-mini"),
+				disableExtensionDiscovery: true,
+				skills: [],
+				contextFiles: [],
+				promptTemplates: [],
+				slashCommands: [],
+				enableMCP: false,
+				enableLsp: false,
+			});
+			session = result.session;
+
 			expect(session.isAdvisorActive()).toBe(true);
 			const dump = session.formatAdvisorHistoryAsText();
 			expect(dump).not.toBeNull();
@@ -67,14 +69,18 @@ describe("advisor watchdog prompt discovery", () => {
 			expect(dump).toContain(watchdogContent);
 			expect(dump).toContain("</attention>");
 		} finally {
-			await session.dispose();
+			try {
+				await session?.dispose();
+			} finally {
+				authStorage.close();
+			}
 		}
 	});
 
 	it("resolves nested folders and sorts by depth", async () => {
-		const tempDir = fs.mkdtempSync(path.join(os.tmpdir(), `pi-advisor-watchdog-${Snowflake.next()}-`));
+		const tempDir = TempDir.createSync("@pi-advisor-watchdog-");
 		tempDirs.push(tempDir);
-		const parentCwd = path.join(tempDir, "project-root");
+		const parentCwd = tempDir.join("project-root");
 		const childCwd = path.join(parentCwd, "subfolder");
 		fs.mkdirSync(childCwd, { recursive: true });
 
@@ -84,36 +90,37 @@ describe("advisor watchdog prompt discovery", () => {
 		fs.writeFileSync(path.join(parentCwd, "WATCHDOG.md"), parentWatchdogContent, "utf8");
 		fs.writeFileSync(path.join(childCwd, "WATCHDOG.md"), childWatchdogContent, "utf8");
 
-		const authStorage = await AuthStorage.create(path.join(tempDir, "testauth.db"));
-		authStorage.setRuntimeApiKey("openai", "test-key");
-		const modelRegistry = new ModelRegistry(authStorage);
-
-		const sessionManager = SessionManager.create(childCwd, path.join(tempDir, "sessions"));
-		const { session } = await createAgentSession({
-			cwd: childCwd,
-			agentDir: tempDir,
-			sessionManager,
-			authStorage,
-			modelRegistry,
-			settings: (() => {
-				const s = Settings.isolated({
-					"async.enabled": false,
-					"advisor.enabled": true,
-				});
-				s.setModelRole("advisor", "openai/gpt-4o-mini");
-				return s;
-			})(),
-			model: getBundledModel("openai", "gpt-4o-mini"),
-			disableExtensionDiscovery: true,
-			skills: [],
-			contextFiles: [],
-			promptTemplates: [],
-			slashCommands: [],
-			enableMCP: false,
-			enableLsp: false,
-		});
-
+		const authStorage = await AuthStorage.create(tempDir.join("testauth.db"));
+		let session: AgentSession | undefined;
 		try {
+			authStorage.setRuntimeApiKey("openai", "test-key");
+			const modelRegistry = new ModelRegistry(authStorage);
+			const sessionManager = SessionManager.create(childCwd, tempDir.join("sessions"));
+			const result = await createAgentSession({
+				cwd: childCwd,
+				agentDir: tempDir.path(),
+				sessionManager,
+				authStorage,
+				modelRegistry,
+				settings: (() => {
+					const s = Settings.isolated({
+						"async.enabled": false,
+						"advisor.enabled": true,
+					});
+					s.setModelRole("advisor", "openai/gpt-4o-mini");
+					return s;
+				})(),
+				model: getBundledModel("openai", "gpt-4o-mini"),
+				disableExtensionDiscovery: true,
+				skills: [],
+				contextFiles: [],
+				promptTemplates: [],
+				slashCommands: [],
+				enableMCP: false,
+				enableLsp: false,
+			});
+			session = result.session;
+
 			expect(session.isAdvisorActive()).toBe(true);
 			const dump = session.formatAdvisorHistoryAsText();
 			expect(dump).not.toBeNull();
@@ -130,16 +137,20 @@ describe("advisor watchdog prompt discovery", () => {
 			expect(childIndex).toBeGreaterThan(-1);
 			expect(parentIndex).toBeLessThan(childIndex);
 		} finally {
-			await session.dispose();
+			try {
+				await session?.dispose();
+			} finally {
+				authStorage.close();
+			}
 		}
 	});
 
 	it("discovers user-level and native project-level watchdog files", async () => {
-		const tempDir = fs.mkdtempSync(path.join(os.tmpdir(), `pi-advisor-watchdog-${Snowflake.next()}-`));
+		const tempDir = TempDir.createSync("@pi-advisor-watchdog-");
 		tempDirs.push(tempDir);
-		const cwd = path.join(tempDir, "project-root");
+		const cwd = tempDir.join("project-root");
 		const ompDir = path.join(cwd, ".omp");
-		const userAgentDir = path.join(tempDir, "user-agent");
+		const userAgentDir = tempDir.join("user-agent");
 		fs.mkdirSync(cwd, { recursive: true });
 		fs.mkdirSync(ompDir, { recursive: true });
 		fs.mkdirSync(userAgentDir, { recursive: true });
@@ -152,36 +163,37 @@ describe("advisor watchdog prompt discovery", () => {
 		fs.writeFileSync(path.join(ompDir, "WATCHDOG.md"), nativeWatchdogContent, "utf8");
 		fs.writeFileSync(path.join(cwd, "WATCHDOG.md"), standaloneWatchdogContent, "utf8");
 
-		const authStorage = await AuthStorage.create(path.join(tempDir, "testauth.db"));
-		authStorage.setRuntimeApiKey("openai", "test-key");
-		const modelRegistry = new ModelRegistry(authStorage);
-
-		const sessionManager = SessionManager.create(cwd, path.join(tempDir, "sessions"));
-		const { session } = await createAgentSession({
-			cwd,
-			agentDir: userAgentDir,
-			sessionManager,
-			authStorage,
-			modelRegistry,
-			settings: (() => {
-				const s = Settings.isolated({
-					"async.enabled": false,
-					"advisor.enabled": true,
-				});
-				s.setModelRole("advisor", "openai/gpt-4o-mini");
-				return s;
-			})(),
-			model: getBundledModel("openai", "gpt-4o-mini"),
-			disableExtensionDiscovery: true,
-			skills: [],
-			contextFiles: [],
-			promptTemplates: [],
-			slashCommands: [],
-			enableMCP: false,
-			enableLsp: false,
-		});
-
+		const authStorage = await AuthStorage.create(tempDir.join("testauth.db"));
+		let session: AgentSession | undefined;
 		try {
+			authStorage.setRuntimeApiKey("openai", "test-key");
+			const modelRegistry = new ModelRegistry(authStorage);
+			const sessionManager = SessionManager.create(cwd, tempDir.join("sessions"));
+			const result = await createAgentSession({
+				cwd,
+				agentDir: userAgentDir,
+				sessionManager,
+				authStorage,
+				modelRegistry,
+				settings: (() => {
+					const s = Settings.isolated({
+						"async.enabled": false,
+						"advisor.enabled": true,
+					});
+					s.setModelRole("advisor", "openai/gpt-4o-mini");
+					return s;
+				})(),
+				model: getBundledModel("openai", "gpt-4o-mini"),
+				disableExtensionDiscovery: true,
+				skills: [],
+				contextFiles: [],
+				promptTemplates: [],
+				slashCommands: [],
+				enableMCP: false,
+				enableLsp: false,
+			});
+			session = result.session;
+
 			expect(session.isAdvisorActive()).toBe(true);
 			const dump = session.formatAdvisorHistoryAsText();
 			expect(dump).not.toBeNull();
@@ -204,7 +216,11 @@ describe("advisor watchdog prompt discovery", () => {
 			expect(userIndex).toBeLessThan(nativeIndex);
 			expect(userIndex).toBeLessThan(standaloneIndex);
 		} finally {
-			await session.dispose();
+			try {
+				await session?.dispose();
+			} finally {
+				authStorage.close();
+			}
 		}
 	});
 });

packages/coding-agent/test/agent-hub-activate.test.ts

@@ -4,6 +4,7 @@
  * focus failure keeps the hub open and surfaces the error as a notice.
  */
 import { afterEach, beforeAll, beforeEach, describe, expect, it } from "bun:test";
+import * as path from "node:path";
 import { resetSettingsForTest, Settings } from "@oh-my-pi/pi-coding-agent/config/settings";
 import { IrcBus } from "@oh-my-pi/pi-coding-agent/irc/bus";
 import { AgentHubOverlayComponent } from "@oh-my-pi/pi-coding-agent/modes/components/agent-hub";
@@ -16,6 +17,7 @@ import type { AgentSession } from "@oh-my-pi/pi-coding-agent/session/agent-sessi
 import { TempDir } from "@oh-my-pi/pi-utils";
 
 const AGENT_ID = "Worker";
+const TEST_CWD = path.resolve("agent-hub-cwd");
 
 function makeHub(focusAgent: (id: string) => Promise<void>) {
 	const agents = new AgentRegistry();
@@ -89,9 +91,10 @@ describe("Agent hub Enter activation", () => {
 
 	it("lists persisted subagent session files after restart", async () => {
 		using tempDir = TempDir.createSync("@omp-agent-hub-persisted-");
-		const sessionFile = `${tempDir.path()}/main.jsonl`;
+		const sessionFile = path.join(tempDir.path(), "main.jsonl");
+		const workerSessionFile = path.join(tempDir.path(), "main", "Worker.jsonl");
 		await Bun.write(sessionFile, "");
-		await Bun.write(`${tempDir.path()}/main/Worker.jsonl`, "");
+		await Bun.write(workerSessionFile, "");
 		const agents = new AgentRegistry();
 		const hub = new AgentHubOverlayComponent({
 			observers: new SessionObserverRegistry(),
@@ -107,7 +110,7 @@ describe("Agent hub Enter activation", () => {
 		const rendered = Bun.stripANSI(hub.render(120).join("\n"));
 		expect(rendered).toContain("Worker");
 		expect(rendered).toContain("parked");
-		expect(agents.get("Worker")?.sessionFile).toBe(`${tempDir.path()}/main/Worker.jsonl`);
+		expect(agents.get("Worker")?.sessionFile).toBe(workerSessionFile);
 		hub.dispose();
 	});
 
@@ -154,7 +157,7 @@ describe("Agent hub Enter activation", () => {
 				focusResolved.resolve();
 			},
 			session: { getToolByName: () => undefined, extensionRunner: undefined },
-			sessionManager: { getCwd: () => "/tmp", getSessionFile: () => null },
+			sessionManager: { getCwd: () => TEST_CWD, getSessionFile: () => null },
 			hideThinkingBlock: false,
 		};
 		const controller = new SelectorController(ctx as unknown as InteractiveModeContext);
@@ -203,7 +206,7 @@ describe("Agent hub double-← gating", () => {
 			collabGuest: { agentRegistry: agents, hubRemote: undefined },
 			focusAgentSession: async () => {},
 			session: { getToolByName: () => undefined, extensionRunner: undefined },
-			sessionManager: { getCwd: () => "/tmp", getSessionFile: () => null },
+			sessionManager: { getCwd: () => TEST_CWD, getSessionFile: () => null },
 			hideThinkingBlock: false,
 		};
 		const controller = new SelectorController(ctx as unknown as InteractiveModeContext);

packages/coding-agent/test/agent-session-advisor-suppression.test.ts

@@ -18,9 +18,6 @@
  *     follow-up stays queued for the next explicit resume rather than auto-running.
  */
 import { afterEach, beforeEach, describe, expect, it } from "bun:test";
-import * as fs from "node:fs";
-import * as os from "node:os";
-import * as path from "node:path";
 import { Agent, type AgentMessage } from "@oh-my-pi/pi-agent-core";
 import { createMockModel, type MockModel, type MockResponse } from "@oh-my-pi/pi-ai/providers/mock";
 import { getBundledModel } from "@oh-my-pi/pi-catalog/models";
@@ -31,7 +28,7 @@ import { AgentSession } from "@oh-my-pi/pi-coding-agent/session/agent-session";
 import { AuthStorage } from "@oh-my-pi/pi-coding-agent/session/auth-storage";
 import { USER_INTERRUPT_LABEL } from "@oh-my-pi/pi-coding-agent/session/messages";
 import { SessionManager } from "@oh-my-pi/pi-coding-agent/session/session-manager";
-import { Snowflake } from "@oh-my-pi/pi-utils";
+import { Snowflake, TempDir } from "@oh-my-pi/pi-utils";
 
 const ADVISOR_TYPE = "advisor";
 
@@ -45,20 +42,23 @@ interface ParkedHarness {
 }
 
 describe("AgentSession advisor auto-resume suppression", () => {
-	let tempDir: string;
+	let tempDir: TempDir;
 	let session: AgentSession;
 	const authStorages: AuthStorage[] = [];
 
 	beforeEach(() => {
-		tempDir = path.join(os.tmpdir(), `pi-advisor-suppress-${Snowflake.next()}`);
-		fs.mkdirSync(tempDir, { recursive: true });
+		tempDir = TempDir.createSync("@pi-advisor-suppress-");
 	});
 
 	afterEach(async () => {
 		// dispose() aborts the agent, cancelling the parked first-turn stream.
-		await session?.dispose();
-		for (const authStorage of authStorages.splice(0)) authStorage.close();
-		fs.rmSync(tempDir, { recursive: true, force: true });
+		try {
+			await session?.dispose();
+		} finally {
+			for (const authStorage of authStorages.splice(0)) authStorage.close();
+			await Bun.sleep(0);
+			await tempDir?.remove();
+		}
 	});
 
 	/**
@@ -86,10 +86,10 @@ describe("AgentSession advisor auto-resume suppression", () => {
 		});
 		const sessionManager = SessionManager.inMemory();
 		const settings = Settings.isolated({ "compaction.enabled": false });
-		const authStorage = await AuthStorage.create(path.join(tempDir, `auth-${Snowflake.next()}.db`));
+		const authStorage = await AuthStorage.create(tempDir.join(`auth-${Snowflake.next()}.db`));
 		authStorages.push(authStorage);
 		authStorage.setRuntimeApiKey("anthropic", "test-key");
-		const modelRegistry = new ModelRegistry(authStorage, path.join(tempDir, "models.yml"));
+		const modelRegistry = new ModelRegistry(authStorage, tempDir.join("models.yml"));
 		session = new AgentSession({ agent, sessionManager, settings, modelRegistry });
 		return { session, sessionManager, mock, streamStarted: started.promise };
 	}

packages/coding-agent/test/agent-session-auto-compaction-queue.test.ts

@@ -125,12 +125,19 @@ describe("AgentSession auto-compaction queue resume", () => {
 	});
 
 	afterEach(async () => {
-		await session.dispose();
-		authStorage.close();
-		tempDir.removeSync();
-		vi.useRealTimers();
-		getRuntimeSignals().length = 0;
-		vi.restoreAllMocks();
+		try {
+			await session?.dispose();
+		} finally {
+			try {
+				authStorage?.close();
+				vi.useRealTimers();
+				await Bun.sleep(0);
+				await tempDir?.remove();
+			} finally {
+				getRuntimeSignals().length = 0;
+				vi.restoreAllMocks();
+			}
+		}
 	});
 
 	it("resumes after threshold compaction when only agent-level queued messages exist", async () => {

packages/coding-agent/test/agent-session-python-cleanup.test.ts

@@ -1,7 +1,5 @@
 import { afterEach, beforeEach, describe, expect, it, type Mock, vi } from "bun:test";
 import * as fs from "node:fs";
-import * as os from "node:os";
-import * as path from "node:path";
 import { getBundledModel } from "@oh-my-pi/pi-catalog/models";
 import { Settings } from "@oh-my-pi/pi-coding-agent/config/settings";
 import * as pythonExecutor from "@oh-my-pi/pi-coding-agent/eval/py/executor";
@@ -9,8 +7,9 @@ import type { PythonKernel as PythonKernelInstance } from "@oh-my-pi/pi-coding-a
 import * as pythonKernel from "@oh-my-pi/pi-coding-agent/eval/py/kernel";
 import { AgentRegistry } from "@oh-my-pi/pi-coding-agent/registry/agent-registry";
 import { createAgentSession, type ExtensionFactory, type WorkspaceTree } from "@oh-my-pi/pi-coding-agent/sdk";
+import { AgentStorage } from "@oh-my-pi/pi-coding-agent/session/agent-storage";
 import { SessionManager } from "@oh-my-pi/pi-coding-agent/session/session-manager";
-import { Snowflake } from "@oh-my-pi/pi-utils";
+import { Snowflake, TempDir } from "@oh-my-pi/pi-utils";
 
 const OK_EXECUTION = { status: "ok", cancelled: false, timedOut: false, stdinRequested: false } as const;
 
@@ -69,12 +68,21 @@ const getModel = () => {
 };
 
 const createTempProject = () => {
-	const tempDir = fs.mkdtempSync(path.join(os.tmpdir(), `pi-agent-session-python-cleanup-${Snowflake.next()}-`));
-	const cwd = path.join(tempDir, "project");
+	const tempDir = TempDir.createSync(`@pi-agent-session-python-cleanup-${Snowflake.next()}-`);
+	const cwd = tempDir.join("project");
 	fs.mkdirSync(cwd, { recursive: true });
 	return { tempDir, cwd };
 };
 
+// createAgentSession opens an AuthStorage at <agentDir>/auth.db that is not
+// closed when construction fails. Point agentDir at a separate dir so the
+// auth.db handle doesn't keep the per-test project temp dir locked on Windows.
+const agentDirPool: TempDir[] = [];
+const createAgentDir = (): string => {
+	const dir = TempDir.createSync("@pi-python-cleanup-agentdir-");
+	agentDirPool.push(dir);
+	return dir.path();
+};
 const emptyWorkspaceTree = (cwd: string): WorkspaceTree => ({
 	rootPath: cwd,
 	rendered: ".",
@@ -102,14 +110,14 @@ const expectSleepNear = (sleepSpy: Mock<typeof Bun.sleep>, targetMs: number) =>
 	).toBe(true);
 };
 const createSession = async (
-	tempDir: string,
+	_tempDir: TempDir,
 	cwd: string,
 	options: { extensions?: ExtensionFactory[]; sessionManager?: SessionManager } = {},
 ) =>
 	(
 		await createAgentSession({
 			cwd,
-			agentDir: tempDir,
+			agentDir: createAgentDir(),
 			sessionManager: options.sessionManager ?? SessionManager.inMemory(cwd),
 			settings: Settings.isolated({ "python.kernelMode": "session" }),
 			model: getModel(),
@@ -126,7 +134,6 @@ const createSession = async (
 			toolNames: ["eval"],
 		})
 	).session;
-
 const createMockKernel = () => {
 	let alive = true;
 	return {
@@ -144,7 +151,7 @@ const createMockKernel = () => {
 };
 
 describe("AgentSession python cleanup", () => {
-	const tempDirs: string[] = [];
+	const tempDirs: TempDir[] = [];
 	let originalNullPrompt: string | undefined;
 
 	beforeEach(() => {
@@ -160,9 +167,15 @@ describe("AgentSession python cleanup", () => {
 		}
 		originalNullPrompt = undefined;
 		vi.restoreAllMocks();
+		AgentStorage.resetInstance();
 		await pythonExecutor.disposeAllKernelSessions();
-		for (const tempDir of tempDirs.splice(0)) {
-			fs.rmSync(tempDir, { recursive: true, force: true });
+		await Bun.sleep(0);
+		// Best-effort cleanup: createAgentSession opens AuthStorage/AgentStorage
+		// inside agentDir that may outlive the test (dispose() doesn't close them).
+		// On Windows the leaked SQLite handles keep the dir locked; swallow EBUSY
+		// rather than failing the test — the OS temp dir reaper will clean up.
+		for (const tempDir of [...tempDirs.splice(0), ...agentDirPool.splice(0)]) {
+			await tempDir.remove().catch(() => {});
 		}
 	});
 
@@ -170,7 +183,7 @@ describe("AgentSession python cleanup", () => {
 		const { tempDir, cwd } = createTempProject();
 		tempDirs.push(tempDir);
 		const unrelatedKernel = createMockKernel();
-		const unrelatedCwd = path.join(tempDir, "unrelated-before");
+		const unrelatedCwd = tempDir.join("unrelated-before");
 		const throwingExtension: ExtensionFactory = () => {
 			throw new Error("Extension init failed");
 		};
@@ -189,7 +202,7 @@ describe("AgentSession python cleanup", () => {
 		await expect(
 			createAgentSession({
 				cwd,
-				agentDir: tempDir,
+				agentDir: createAgentDir(),
 				sessionManager: SessionManager.inMemory(cwd),
 				settings: Settings.isolated({ "python.kernelMode": "session" }),
 				model: getModel(),
@@ -237,7 +250,7 @@ describe("AgentSession python cleanup", () => {
 		const { tempDir, cwd } = createTempProject();
 		tempDirs.push(tempDir);
 		const unrelatedKernel = createMockKernel();
-		const unrelatedCwd = path.join(tempDir, "unrelated-after");
+		const unrelatedCwd = tempDir.join("unrelated-after");
 		vi.spyOn(pythonKernel, "checkPythonKernelAvailability").mockResolvedValue({ ok: true });
 		const startSpy = vi
 			.spyOn(pythonKernel.PythonKernel, "start")
@@ -257,7 +270,7 @@ describe("AgentSession python cleanup", () => {
 		await expect(
 			createAgentSession({
 				cwd,
-				agentDir: tempDir,
+				agentDir: createAgentDir(),
 				sessionManager: SessionManager.inMemory(cwd),
 				settings: Settings.isolated({ "python.kernelMode": "session", "memory.backend": "local" }),
 				model: getModel(),

packages/coding-agent/test/agent-storage-sqlite-compat.test.ts

@@ -1,9 +1,8 @@
 import { Database } from "bun:sqlite";
 import { afterEach, describe, expect, it } from "bun:test";
-import * as fs from "node:fs/promises";
-import * as os from "node:os";
 import * as path from "node:path";
 import { AgentStorage } from "@oh-my-pi/pi-coding-agent/session/agent-storage";
+import { TempDir } from "@oh-my-pi/pi-utils";
 import { readTableSql } from "./helpers/sqlite-inspect";
 
 const LEGACY_TIMESTAMP = 1_700_000_000;
@@ -34,18 +33,21 @@ function readSettingsRows(dbPath: string): Array<{ key: string; value: string; u
 }
 
 describe("AgentStorage SQLite compatibility", () => {
-	let tempDir = "";
+	let tempDir: TempDir;
 
 	afterEach(async () => {
+		AgentStorage.resetInstance();
 		if (tempDir) {
-			await fs.rm(tempDir, { recursive: true, force: true });
-			tempDir = "";
+			try {
+				await tempDir.remove();
+			} catch {}
+			tempDir = undefined as unknown as TempDir;
 		}
 	});
 
 	it("creates fresh storage without unixepoch defaults", async () => {
-		tempDir = await fs.mkdtemp(path.join(os.tmpdir(), "omp-agent-storage-fresh-"));
-		const dbPath = path.join(tempDir, "agent.db");
+		tempDir = TempDir.createSync("omp-agent-storage-fresh-");
+		const dbPath = path.join(tempDir.path(), "agent.db");
 
 		const storage = await AgentStorage.open(dbPath);
 		storage.recordModelUsage("openai/gpt-5");
@@ -59,8 +61,8 @@ describe("AgentStorage SQLite compatibility", () => {
 	});
 
 	it("migrates legacy settings and model usage schemas away from unixepoch defaults", async () => {
-		tempDir = await fs.mkdtemp(path.join(os.tmpdir(), "omp-agent-storage-legacy-"));
-		const dbPath = path.join(tempDir, "agent.db");
+		tempDir = TempDir.createSync("omp-agent-storage-legacy-");
+		const dbPath = path.join(tempDir.path(), "agent.db");
 		const legacyDb = new Database(dbPath);
 		legacyDb.exec(`
 			CREATE TABLE schema_version (version INTEGER PRIMARY KEY);

packages/coding-agent/test/autocomplete-max-visible.test.ts

@@ -1,35 +1,38 @@
 import { afterEach, beforeEach, describe, expect, it, vi } from "bun:test";
 import * as fs from "node:fs";
-import * as os from "node:os";
 import * as path from "node:path";
 import { resetSettingsForTest, Settings } from "@oh-my-pi/pi-coding-agent/config/settings";
 import { SelectorController } from "@oh-my-pi/pi-coding-agent/modes/controllers/selector-controller";
-import { getProjectAgentDir, Snowflake } from "@oh-my-pi/pi-utils";
+import { AgentStorage } from "@oh-my-pi/pi-coding-agent/session/agent-storage";
+import { getProjectAgentDir, TempDir } from "@oh-my-pi/pi-utils";
 import { YAML } from "bun";
 import { beginSettingsTest, restoreSettingsTestState, type SettingsTestState } from "./helpers/settings-test-state";
 
 describe("autocompleteMaxVisible setting", () => {
 	let settingsState: SettingsTestState | undefined;
-	let testDir = "";
+	let tempDir: TempDir;
 	let agentDir: string;
 	let projectDir: string;
 
 	beforeEach(() => {
 		settingsState = beginSettingsTest();
-		testDir = path.join(os.tmpdir(), "test-autocomplete-settings", Snowflake.next());
-		agentDir = path.join(testDir, "agent");
-		projectDir = path.join(testDir, "project");
+		tempDir = TempDir.createSync("test-autocomplete-settings-");
+		agentDir = path.join(tempDir.path(), "agent");
+		projectDir = path.join(tempDir.path(), "project");
 		fs.mkdirSync(agentDir, { recursive: true });
 		fs.mkdirSync(getProjectAgentDir(projectDir), { recursive: true });
 	});
 
-	afterEach(() => {
+	afterEach(async () => {
+		AgentStorage.resetInstance();
 		restoreSettingsTestState(settingsState);
 		settingsState = undefined;
-		if (testDir && fs.existsSync(testDir)) {
-			fs.rmSync(testDir, { recursive: true, force: true });
+		if (tempDir) {
+			try {
+				await tempDir.remove();
+			} catch {}
+			tempDir = undefined as unknown as TempDir;
 		}
-		testDir = "";
 	});
 
 	it("should persist and read back a configured value", async () => {

packages/coding-agent/test/autoresearch-state.test.ts

@@ -1,7 +1,4 @@
 import { afterEach, beforeEach, describe, expect, it, vi } from "bun:test";
-import * as fs from "node:fs";
-import * as os from "node:os";
-import * as path from "node:path";
 import { createAutoresearchExtension } from "@oh-my-pi/pi-coding-agent/autoresearch";
 import {
 	buildExperimentState,
@@ -11,7 +8,7 @@ import {
 	findBestKeptMetric,
 	reconstructControlState,
 } from "@oh-my-pi/pi-coding-agent/autoresearch/state";
-import { AutoresearchStorage } from "@oh-my-pi/pi-coding-agent/autoresearch/storage";
+import { AutoresearchStorage, closeAllAutoresearchStorages } from "@oh-my-pi/pi-coding-agent/autoresearch/storage";
 import type { ExperimentResult } from "@oh-my-pi/pi-coding-agent/autoresearch/types";
 import type {
 	ExtensionAPI,
@@ -19,16 +16,14 @@ import type {
 	RegisteredCommand,
 } from "@oh-my-pi/pi-coding-agent/extensibility/extensions";
 import * as git from "@oh-my-pi/pi-coding-agent/utils/git";
-import { Snowflake } from "@oh-my-pi/pi-utils";
+import { TempDir } from "@oh-my-pi/pi-utils";
 
 afterEach(() => {
 	vi.restoreAllMocks();
 });
 
-function makeTempDir(): string {
-	const dir = path.join(os.tmpdir(), `pi-autoresearch-test-${Snowflake.next()}`);
-	fs.mkdirSync(dir, { recursive: true });
-	return dir;
+function makeTempDir(): TempDir {
+	return TempDir.createSync("@pi-autoresearch-test-");
 }
 
 function makeResult(partial: Partial<ExperimentResult>): ExperimentResult {
@@ -113,18 +108,19 @@ describe("autoresearch state math", () => {
 });
 
 describe("AutoresearchStorage round-trip", () => {
-	let dbDir: string;
+	let dbDir: TempDir;
 
 	beforeEach(() => {
 		dbDir = makeTempDir();
 	});
 
-	afterEach(() => {
-		fs.rmSync(dbDir, { recursive: true, force: true });
+	afterEach(async () => {
+		await Bun.sleep(0);
+		await dbDir.remove().catch(() => {});
 	});
 
 	function openStorage(): AutoresearchStorage {
-		return new AutoresearchStorage(path.join(dbDir, "test.db"), dbDir);
+		return new AutoresearchStorage(dbDir.join("test.db"), dbDir.path());
 	}
 
 	it("persists sessions and exposes the active session", () => {
@@ -516,25 +512,25 @@ function createCommandHarness(
 }
 
 describe("autoresearch slash command", () => {
-	const cleanups: string[] = [];
-	let dbOverride: string | undefined;
+	const cleanups: TempDir[] = [];
+	let dbOverride: TempDir | undefined;
 
 	beforeEach(() => {
-		dbOverride = path.join(os.tmpdir(), `pi-autoresearch-cmd-${Snowflake.next()}`);
-		fs.mkdirSync(dbOverride, { recursive: true });
-		process.env.OMP_AUTORESEARCH_DB_DIR = dbOverride;
+		dbOverride = TempDir.createSync("@pi-autoresearch-cmd-");
+		process.env.OMP_AUTORESEARCH_DB_DIR = dbOverride.path();
 		cleanups.push(dbOverride);
 	});
 
 	afterEach(() => {
 		delete process.env.OMP_AUTORESEARCH_DB_DIR;
+		closeAllAutoresearchStorages();
 		for (const dir of cleanups.splice(0)) {
-			fs.rmSync(dir, { recursive: true, force: true });
+			dir.removeSync();
 		}
 	});
 
 	it("enables autoresearch with a notify when invoked bare in a clean repo", async () => {
-		const dir = makeTempDir();
+		const dir = makeTempDir().path();
 		const harness = createCommandHarness(dir, async (_command, args) => {
 			if (args[0] === "rev-parse") return { code: 0, stderr: "", stdout: `${dir}\n` };
 			if (args[0] === "branch" && args[1] === "--show-current") return { code: 0, stderr: "", stdout: "main\n" };
@@ -549,7 +545,7 @@ describe("autoresearch slash command", () => {
 	});
 
 	it("forwards a slash argument as the user message and creates a slug branch", async () => {
-		const dir = makeTempDir();
+		const dir = makeTempDir().path();
 		const harness = createCommandHarness(dir, async (_command, args) => {
 			if (args[0] === "rev-parse") return { code: 0, stderr: "", stdout: `${dir}\n` };
 			if (args[0] === "branch" && args[1] === "--show-current") return { code: 0, stderr: "", stdout: "main\n" };
@@ -565,7 +561,7 @@ describe("autoresearch slash command", () => {
 	});
 
 	it("aborts with an error when the worktree is dirty", async () => {
-		const dir = makeTempDir();
+		const dir = makeTempDir().path();
 		const harness = createCommandHarness(dir, async (_command, args) => {
 			if (args[0] === "rev-parse") return { code: 0, stderr: "", stdout: `${dir}\n` };
 			if (args[0] === "branch" && args[1] === "--show-current") return { code: 0, stderr: "", stdout: "main\n" };
@@ -583,20 +579,20 @@ describe("autoresearch slash command", () => {
 });
 
 describe("autoresearch tool-call hook", () => {
-	const cleanups: string[] = [];
-	let dbOverride: string;
+	const cleanups: TempDir[] = [];
+	let dbOverride: TempDir;
 
 	beforeEach(() => {
-		dbOverride = path.join(os.tmpdir(), `pi-autoresearch-hook-${Snowflake.next()}`);
-		fs.mkdirSync(dbOverride, { recursive: true });
-		process.env.OMP_AUTORESEARCH_DB_DIR = dbOverride;
+		dbOverride = TempDir.createSync("@pi-autoresearch-hook-");
+		process.env.OMP_AUTORESEARCH_DB_DIR = dbOverride.path();
 		cleanups.push(dbOverride);
 	});
 
 	afterEach(() => {
 		delete process.env.OMP_AUTORESEARCH_DB_DIR;
+		closeAllAutoresearchStorages();
 		for (const dir of cleanups.splice(0)) {
-			fs.rmSync(dir, { recursive: true, force: true });
+			dir.removeSync();
 		}
 	});
 

packages/coding-agent/test/autoresearch-tools.test.ts

@@ -1,11 +1,11 @@
 import { afterAll, afterEach, beforeAll, beforeEach, describe, expect, it, vi } from "bun:test";
 import * as fs from "node:fs";
-import * as os from "node:os";
 import * as path from "node:path";
 import type { ImageContent, TextContent } from "@oh-my-pi/pi-ai";
 import { createSessionRuntime } from "@oh-my-pi/pi-coding-agent/autoresearch/state";
 import {
 	type AutoresearchStorage,
+	closeAllAutoresearchStorages,
 	openAutoresearchStorage,
 	type SessionRow,
 } from "@oh-my-pi/pi-coding-agent/autoresearch/storage";
@@ -16,7 +16,7 @@ import { createUpdateNotesTool } from "@oh-my-pi/pi-coding-agent/autoresearch/to
 import type { ASIData, LogDetails, NumericMetricMap, RunDetails } from "@oh-my-pi/pi-coding-agent/autoresearch/types";
 import type { ExtensionAPI, ExtensionContext } from "@oh-my-pi/pi-coding-agent/extensibility/extensions";
 import * as git from "@oh-my-pi/pi-coding-agent/utils/git";
-import { Snowflake } from "@oh-my-pi/pi-utils";
+import { TempDir } from "@oh-my-pi/pi-utils";
 import { $ } from "bun";
 
 afterEach(() => {
@@ -29,10 +29,8 @@ function firstTextBlockText(content: Array<TextContent | ImageContent>): string
 	return block.text;
 }
 
-function makeTempDir(prefix = "pi-autoresearch-tools"): string {
-	const dir = path.join(os.tmpdir(), `${prefix}-${Snowflake.next()}`);
-	fs.mkdirSync(dir, { recursive: true });
-	return dir;
+function makeTempDir(prefix = "@pi-autoresearch-tools-"): TempDir {
+	return TempDir.createSync(prefix);
 }
 
 function dashboardStub() {
@@ -78,43 +76,47 @@ function createPiHarness(initialTools: string[] = []): PiHarness {
 // inside a real repo, so the production tools resolve HEAD/branch from `.git` on
 // disk (sub-millisecond) instead of spawning fallback git subprocesses for every
 // `repo.root` / `branch.current` / `head.sha` lookup against a bare temp dir.
-let templateRepo: string;
-let templateBranchRepo: string;
+let templateRepo: TempDir;
+let templateBranchRepo: TempDir;
 let templateBaselineCommit: string;
 
 beforeAll(async () => {
-	templateRepo = makeTempDir("pi-autoresearch-template");
-	await Bun.write(path.join(templateRepo, "README.md"), "# baseline\n");
-	await $`git init --initial-branch=main && git config user.email tester@example.com && git config user.name Tester && git add -A && git commit -m baseline`
-		.cwd(templateRepo)
+	templateRepo = makeTempDir("@pi-autoresearch-template-");
+	await Bun.write(path.join(templateRepo.path(), "README.md"), "# baseline\n");
+	await $`git init --initial-branch=main && git config core.autocrlf false && git config user.email tester@example.com && git config user.name Tester && git add -A && git commit -m baseline`
+		.cwd(templateRepo.path())
 		.quiet();
-	templateBaselineCommit = (await $`git rev-parse HEAD`.cwd(templateRepo).text()).trim();
+	templateBaselineCommit = (await $`git rev-parse HEAD`.cwd(templateRepo.path()).text()).trim();
 	// Second fixture: harness committed and already on an `autoresearch/*` branch,
 	// the baseline for log_experiment's on-branch keep/discard scenarios.
-	templateBranchRepo = makeTempDir("pi-autoresearch-template-branch");
-	fs.cpSync(templateRepo, templateBranchRepo, { recursive: true });
-	await Bun.write(path.join(templateBranchRepo, "autoresearch.sh"), "#!/usr/bin/env bash\necho METRIC m=1\n");
-	await $`git add -A && git commit -m harness && git checkout -b autoresearch/base`.cwd(templateBranchRepo).quiet();
+	templateBranchRepo = makeTempDir("@pi-autoresearch-template-branch-");
+	fs.cpSync(templateRepo.path(), templateBranchRepo.path(), { recursive: true });
+	await Bun.write(path.join(templateBranchRepo.path(), "autoresearch.sh"), "#!/usr/bin/env bash\necho METRIC m=1\n");
+	await $`git add -A && git commit -m harness && git checkout -b autoresearch/base`
+		.cwd(templateBranchRepo.path())
+		.quiet();
 });
 
-afterAll(() => {
-	fs.rmSync(templateRepo, { recursive: true, force: true });
-	fs.rmSync(templateBranchRepo, { recursive: true, force: true });
+afterAll(async () => {
+	closeAllAutoresearchStorages();
+	await Bun.sleep(0);
+	await templateRepo.remove();
+	await templateBranchRepo.remove();
 });
 
 // Independent working copy of the template repo: baseline commit on `main`,
 // committer identity configured, ready for per-test branch/commit scenarios.
 function freshRepo(): { dir: string; baselineCommit: string } {
-	const dir = makeTempDir();
-	fs.cpSync(templateRepo, dir, { recursive: true });
+	const dir = makeTempDir().path();
+	fs.cpSync(templateRepo.path(), dir, { recursive: true });
 	return { dir, baselineCommit: templateBaselineCommit };
 }
 
 // Like freshRepo, but already on an `autoresearch/*` branch with the harness
 // committed — the baseline for log_experiment's on-branch keep/discard paths.
 function freshBranchRepo(): { dir: string } {
-	const dir = makeTempDir();
-	fs.cpSync(templateBranchRepo, dir, { recursive: true });
+	const dir = makeTempDir().path();
+	fs.cpSync(templateBranchRepo.path(), dir, { recursive: true });
 	return { dir };
 }
 
@@ -162,16 +164,18 @@ function seedCompletedRun(
 }
 
 describe("init_experiment", () => {
-	let dbOverride: string;
+	let dbOverride: TempDir;
 
 	beforeEach(() => {
-		dbOverride = makeTempDir("pi-autoresearch-init-db");
-		process.env.OMP_AUTORESEARCH_DB_DIR = dbOverride;
+		dbOverride = makeTempDir("@pi-autoresearch-init-db-");
+		process.env.OMP_AUTORESEARCH_DB_DIR = dbOverride.path();
 	});
 
-	afterEach(() => {
+	afterEach(async () => {
 		delete process.env.OMP_AUTORESEARCH_DB_DIR;
-		fs.rmSync(dbOverride, { recursive: true, force: true });
+		closeAllAutoresearchStorages();
+		await Bun.sleep(0);
+		await dbOverride.remove();
 	});
 
 	it("opens a new session and persists scope and metric metadata", async () => {
@@ -340,16 +344,18 @@ describe("init_experiment", () => {
 });
 
 describe("run_experiment", () => {
-	let dbOverride: string;
+	let dbOverride: TempDir;
 
 	beforeEach(() => {
-		dbOverride = makeTempDir("pi-autoresearch-run-db");
-		process.env.OMP_AUTORESEARCH_DB_DIR = dbOverride;
+		dbOverride = makeTempDir("@pi-autoresearch-run-db-");
+		process.env.OMP_AUTORESEARCH_DB_DIR = dbOverride.path();
 	});
 
-	afterEach(() => {
+	afterEach(async () => {
 		delete process.env.OMP_AUTORESEARCH_DB_DIR;
-		fs.rmSync(dbOverride, { recursive: true, force: true });
+		closeAllAutoresearchStorages();
+		await Bun.sleep(0);
+		await dbOverride.remove();
 	});
 
 	it("rejects when no session is active", async () => {
@@ -429,16 +435,18 @@ describe("run_experiment", () => {
 });
 
 describe("log_experiment", () => {
-	let dbOverride: string;
+	let dbOverride: TempDir;
 
 	beforeEach(() => {
-		dbOverride = makeTempDir("pi-autoresearch-log-db");
-		process.env.OMP_AUTORESEARCH_DB_DIR = dbOverride;
+		dbOverride = makeTempDir("@pi-autoresearch-log-db-");
+		process.env.OMP_AUTORESEARCH_DB_DIR = dbOverride.path();
 	});
 
-	afterEach(() => {
+	afterEach(async () => {
 		delete process.env.OMP_AUTORESEARCH_DB_DIR;
-		fs.rmSync(dbOverride, { recursive: true, force: true });
+		closeAllAutoresearchStorages();
+		await Bun.sleep(0);
+		await dbOverride.remove();
 	});
 
 	async function setupRun(dir: string, runtime = createSessionRuntime()) {
@@ -563,7 +571,7 @@ describe("log_experiment", () => {
 	it("flags previously logged runs via flag_runs", async () => {
 		// Bare temp dir (no repo): the session is created with `branch: null`, so the
 		// tool's branch lookup must also resolve to null to match it.
-		const dir = makeTempDir();
+		const dir = makeTempDir().path();
 		const storage = await openAutoresearchStorage(dir);
 		const session = storage.openSession({
 			name: "speed",
@@ -831,16 +839,18 @@ describe("log_experiment", () => {
 });
 
 describe("update_notes", () => {
-	let dbOverride: string;
+	let dbOverride: TempDir;
 
 	beforeEach(() => {
-		dbOverride = makeTempDir("pi-autoresearch-notes-db");
-		process.env.OMP_AUTORESEARCH_DB_DIR = dbOverride;
+		dbOverride = makeTempDir("@pi-autoresearch-notes-db-");
+		process.env.OMP_AUTORESEARCH_DB_DIR = dbOverride.path();
 	});
 
-	afterEach(() => {
+	afterEach(async () => {
 		delete process.env.OMP_AUTORESEARCH_DB_DIR;
-		fs.rmSync(dbOverride, { recursive: true, force: true });
+		closeAllAutoresearchStorages();
+		await Bun.sleep(0);
+		await dbOverride.remove().catch(() => {});
 	});
 
 	it("replaces session notes and refreshes runtime state", async () => {

packages/coding-agent/test/config-cli.test.ts

@@ -1,23 +1,23 @@
 import { afterEach, beforeEach, describe, expect, it, vi } from "bun:test";
-import * as fs from "node:fs/promises";
-import * as os from "node:os";
 import * as path from "node:path";
 import { runConfigCommand } from "@oh-my-pi/pi-coding-agent/cli/config-cli";
 import { resetSettingsForTest } from "@oh-my-pi/pi-coding-agent/config/settings";
-import { getConfigRootDir, setAgentDir } from "@oh-my-pi/pi-utils";
+import { AgentStorage } from "@oh-my-pi/pi-coding-agent/session/agent-storage";
+import { getConfigRootDir, setAgentDir, TempDir } from "@oh-my-pi/pi-utils";
 
-let testAgentDir = "";
+let testAgentDir: TempDir | undefined;
 const originalAgentDir = process.env.PI_CODING_AGENT_DIR;
 const fallbackAgentDir = path.join(getConfigRootDir(), "agent");
 
-beforeEach(async () => {
+beforeEach(() => {
 	resetSettingsForTest();
-	testAgentDir = await fs.mkdtemp(path.join(os.tmpdir(), "omp-config-cli-"));
-	setAgentDir(testAgentDir);
+	testAgentDir = TempDir.createSync("omp-config-cli-");
+	setAgentDir(testAgentDir.path());
 });
 
 afterEach(async () => {
 	vi.restoreAllMocks();
+	AgentStorage.resetInstance();
 	resetSettingsForTest();
 	if (originalAgentDir) {
 		setAgentDir(originalAgentDir);
@@ -25,7 +25,12 @@ afterEach(async () => {
 		setAgentDir(fallbackAgentDir);
 		delete process.env.PI_CODING_AGENT_DIR;
 	}
-	await fs.rm(testAgentDir, { recursive: true, force: true });
+	if (testAgentDir) {
+		try {
+			await testAgentDir.remove();
+		} catch {}
+		testAgentDir = undefined;
+	}
 });
 
 describe("config CLI schema coverage", () => {

packages/coding-agent/test/extensions-discovery.test.ts

@@ -26,8 +26,8 @@ describe("extensions discovery", () => {
 		const result = await discoverAndLoadExtensions(configuredPaths, tempDir.path());
 		return {
 			...result,
-			extensions: filterUserScoped(result.extensions),
-			errors: filterUserScoped(result.errors),
+			extensions: filterUserScoped(result.extensions, [tempDir.path(), ...configuredPaths]),
+			errors: filterUserScoped(result.errors, [tempDir.path(), ...configuredPaths]),
 		};
 	};
 
@@ -153,7 +153,7 @@ describe("extensions discovery", () => {
 
 		expect(result.errors).toHaveLength(0);
 		expect(result.extensions).toHaveLength(1);
-		expect(result.extensions[0].path).toContain("linked-package/src/main.ts");
+		expect(result.extensions[0].path).toContain(path.join("linked-package", "src", "main.ts"));
 	});
 
 	it("discovers index.ts in a symlinked extension directory", async () => {
@@ -166,7 +166,7 @@ describe("extensions discovery", () => {
 
 		expect(result.errors).toHaveLength(0);
 		expect(result.extensions).toHaveLength(1);
-		expect(result.extensions[0].path).toContain("linked-index-ts/index.ts");
+		expect(result.extensions[0].path).toContain(path.join("linked-index-ts", "index.ts"));
 	});
 
 	it("discovers index.js in a symlinked extension directory", async () => {
@@ -179,7 +179,7 @@ describe("extensions discovery", () => {
 
 		expect(result.errors).toHaveLength(0);
 		expect(result.extensions).toHaveLength(1);
-		expect(result.extensions[0].path).toContain("linked-index-js/index.js");
+		expect(result.extensions[0].path).toContain(path.join("linked-index-js", "index.js"));
 	});
 
 	it("package.json can declare multiple extensions", async () => {
@@ -442,7 +442,7 @@ describe("extensions discovery", () => {
 
 	it("resolves 3rd party npm dependencies (chalk)", async () => {
 		// Load the real chalk-logger extension from examples
-		const chalkLoggerPath = path.resolve(import.meta.dirname, "../examples/extensions/chalk-logger.ts");
+		const chalkLoggerPath = path.resolve(import.meta.dirname, "..", "examples", "extensions", "chalk-logger.ts");
 
 		const result = await discoverForTest([chalkLoggerPath]);
 

packages/coding-agent/test/history-storage-search.test.ts

@@ -1,15 +1,13 @@
 import { afterEach, beforeEach, describe, expect, it, vi } from "bun:test";
-import * as fs from "node:fs/promises";
-import * as os from "node:os";
-import * as path from "node:path";
 import { HistoryStorage } from "@oh-my-pi/pi-coding-agent/session/history-storage";
+import { TempDir } from "@oh-my-pi/pi-utils";
 
-let tempDir = "";
+let tempDir: TempDir | null = null;
 
 async function freshStorage(): Promise<HistoryStorage> {
-	tempDir = await fs.mkdtemp(path.join(os.tmpdir(), "omp-history-search-"));
+	tempDir = TempDir.createSync("@omp-history-search-");
 	HistoryStorage.resetInstance();
-	return HistoryStorage.open(path.join(tempDir, "history.db"));
+	return HistoryStorage.open(tempDir.join("history.db"));
 }
 
 async function seed(storage: HistoryStorage, prompts: string[]): Promise<void> {
@@ -27,8 +25,9 @@ afterEach(async () => {
 	HistoryStorage.resetInstance();
 	vi.useRealTimers();
 	if (tempDir) {
-		await fs.rm(tempDir, { recursive: true, force: true });
-		tempDir = "";
+		await Bun.sleep(0);
+		await tempDir.remove().catch(() => {});
+		tempDir = null;
 	}
 });
 

packages/coding-agent/test/history-storage-session.test.ts

@@ -1,15 +1,13 @@
 import { Database } from "bun:sqlite";
 import { afterEach, beforeEach, describe, expect, it, vi } from "bun:test";
-import * as fs from "node:fs/promises";
-import * as os from "node:os";
-import * as path from "node:path";
 import { HistoryStorage } from "@oh-my-pi/pi-coding-agent/session/history-storage";
+import { TempDir } from "@oh-my-pi/pi-utils";
 
-let tempDir = "";
+let tempDir: TempDir | null = null;
 
 async function freshStorage(prefix = "omp-history-session-"): Promise<{ storage: HistoryStorage; dbPath: string }> {
-	tempDir = await fs.mkdtemp(path.join(os.tmpdir(), prefix));
-	const dbPath = path.join(tempDir, "history.db");
+	tempDir = TempDir.createSync(`@${prefix}`);
+	const dbPath = tempDir.join("history.db");
 	HistoryStorage.resetInstance();
 	return { storage: HistoryStorage.open(dbPath), dbPath };
 }
@@ -29,8 +27,9 @@ afterEach(async () => {
 	HistoryStorage.resetInstance();
 	vi.useRealTimers();
 	if (tempDir) {
-		await fs.rm(tempDir, { recursive: true, force: true });
-		tempDir = "";
+		await Bun.sleep(0);
+		await tempDir.remove().catch(() => {});
+		tempDir = null;
 	}
 });
 
@@ -84,8 +83,8 @@ describe("HistoryStorage session linkage", () => {
 	});
 
 	it("adds session_id to a pre-existing schema and leaves legacy rows unstamped", async () => {
-		tempDir = await fs.mkdtemp(path.join(os.tmpdir(), "omp-history-session-migrate-"));
-		const dbPath = path.join(tempDir, "history.db");
+		tempDir = TempDir.createSync("@omp-history-session-migrate-");
+		const dbPath = tempDir.join("history.db");
 		const legacyDb = new Database(dbPath);
 		legacyDb.exec(`
 			CREATE TABLE history (

packages/coding-agent/test/history-storage-sqlite-compat.test.ts

@@ -1,14 +1,12 @@
 import { Database } from "bun:sqlite";
 import { afterEach, beforeEach, expect, it } from "bun:test";
-import * as fs from "node:fs/promises";
-import * as os from "node:os";
-import * as path from "node:path";
 import { HistoryStorage } from "@oh-my-pi/pi-coding-agent/session/history-storage";
+import { TempDir } from "@oh-my-pi/pi-utils";
 import { readTableSql } from "./helpers/sqlite-inspect";
 
 const LEGACY_TIMESTAMP = 1_700_000_000;
 
-let tempDir = "";
+let tempDir: TempDir | null = null;
 
 beforeEach(() => {
 	HistoryStorage.resetInstance();
@@ -17,14 +15,15 @@ beforeEach(() => {
 afterEach(async () => {
 	HistoryStorage.resetInstance();
 	if (tempDir) {
-		await fs.rm(tempDir, { recursive: true, force: true });
-		tempDir = "";
+		await Bun.sleep(0);
+		await tempDir.remove().catch(() => {});
+		tempDir = null;
 	}
 });
 
 it("migrates legacy history schema away from unixepoch defaults", async () => {
-	tempDir = await fs.mkdtemp(path.join(os.tmpdir(), "omp-history-storage-legacy-"));
-	const dbPath = path.join(tempDir, "history.db");
+	tempDir = TempDir.createSync("@omp-history-storage-legacy-");
+	const dbPath = tempDir.join("history.db");
 	const legacyDb = new Database(dbPath);
 	legacyDb.exec(`
 		CREATE TABLE history (

packages/coding-agent/test/issue-846-repro.test.ts

@@ -11,14 +11,13 @@
 
 import { afterEach, beforeEach, describe, expect, test, vi } from "bun:test";
 import * as fs from "node:fs/promises";
-import * as os from "node:os";
 import * as path from "node:path";
 import type { Model } from "@oh-my-pi/pi-ai";
 import * as ai from "@oh-my-pi/pi-ai";
 import { Settings } from "@oh-my-pi/pi-coding-agent/config/settings";
 import { startMemoryStartupTask } from "@oh-my-pi/pi-coding-agent/memories";
 import * as memoryStorage from "@oh-my-pi/pi-coding-agent/memories/storage";
-import { getAgentDbPath, logger, Snowflake } from "@oh-my-pi/pi-utils";
+import { getAgentDbPath, logger, Snowflake, TempDir } from "@oh-my-pi/pi-utils";
 
 interface SessionLike {
 	sessionManager: {
@@ -40,13 +39,12 @@ interface ModelRegistryLike {
 	resolver: (...args: unknown[]) => () => Promise<string>;
 }
 
-const createdDirs = new Set<string>();
+const tempDirs: TempDir[] = [];
 
-async function makeTempDir(prefix: string): Promise<string> {
-	const dir = path.join(os.tmpdir(), `${prefix}-${Snowflake.next()}`);
-	await fs.mkdir(dir, { recursive: true });
-	createdDirs.add(dir);
-	return dir;
+function makeTempDir(prefix: string): string {
+	const dir = TempDir.createSync(`@${prefix}-${Snowflake.next()}`);
+	tempDirs.push(dir);
+	return dir.path();
 }
 
 function createModel(): Model {
@@ -83,10 +81,10 @@ describe("issue #846: phase1 stage1 failures must be logged", () => {
 		vi.restoreAllMocks();
 		process.env.XDG_DATA_HOME = savedXdgData;
 		process.env.XDG_STATE_HOME = savedXdgState;
-		for (const dir of createdDirs) {
-			await fs.rm(dir, { recursive: true, force: true });
+		await Bun.sleep(0);
+		for (const dir of tempDirs.splice(0)) {
+			await dir.remove();
 		}
-		createdDirs.clear();
 	});
 
 	test("emits logger.error per failed stage1 claim with the underlying reason", async () => {

packages/coding-agent/test/issue-905-repro.test.ts

@@ -14,20 +14,19 @@
 
 import { afterAll, beforeAll, expect, test } from "bun:test";
 import * as fs from "node:fs/promises";
-import * as os from "node:os";
-import * as path from "node:path";
 import { AuthStorage } from "@oh-my-pi/pi-ai";
 import { runModelsListing } from "@oh-my-pi/pi-coding-agent/cli/models-cli";
 import { ModelRegistry } from "@oh-my-pi/pi-coding-agent/config/model-registry";
+import { TempDir } from "@oh-my-pi/pi-utils";
 
-let tmp: string;
+let tmp: TempDir;
 let extPath: string;
 let dbPath: string;
 
 beforeAll(async () => {
-	tmp = await fs.mkdtemp(path.join(os.tmpdir(), "issue-905-"));
-	extPath = path.join(tmp, "ext.ts");
-	dbPath = path.join(tmp, "auth.db");
+	tmp = await TempDir.create("@issue-905-");
+	extPath = tmp.join("ext.ts");
+	dbPath = tmp.join("auth.db");
 	await fs.writeFile(
 		extPath,
 		`export default function (pi) {
@@ -51,33 +50,38 @@ beforeAll(async () => {
 });
 
 afterAll(async () => {
-	await fs.rm(tmp, { recursive: true, force: true });
+	await Bun.sleep(0);
+	await tmp.remove();
 });
 
 test("omp models surfaces extension-registered providers (issue #905)", async () => {
 	const authStorage = await AuthStorage.create(dbPath);
-	const modelRegistry = new ModelRegistry(authStorage);
-
-	const captured: string[] = [];
-	const originalWrite = process.stdout.write.bind(process.stdout);
-	process.stdout.write = ((chunk: string | Uint8Array) => {
-		captured.push(typeof chunk === "string" ? chunk : Buffer.from(chunk).toString("utf8"));
-		return true;
-	}) as typeof process.stdout.write;
-
 	try {
-		await runModelsListing({
-			modelRegistry,
-			cwd: tmp,
-			action: "ls",
-			additionalExtensionPaths: [extPath],
-			disableExtensionDiscovery: true,
-		});
-	} finally {
-		process.stdout.write = originalWrite;
-	}
+		const modelRegistry = new ModelRegistry(authStorage);
 
-	const output = captured.join("");
-	expect(output).toContain("test-gw");
-	expect(output).toContain("test-model");
+		const captured: string[] = [];
+		const originalWrite = process.stdout.write.bind(process.stdout);
+		process.stdout.write = ((chunk: string | Uint8Array) => {
+			captured.push(typeof chunk === "string" ? chunk : Buffer.from(chunk).toString("utf8"));
+			return true;
+		}) as typeof process.stdout.write;
+
+		try {
+			await runModelsListing({
+				modelRegistry,
+				cwd: tmp.path(),
+				action: "ls",
+				additionalExtensionPaths: [extPath],
+				disableExtensionDiscovery: true,
+			});
+		} finally {
+			process.stdout.write = originalWrite;
+		}
+
+		const output = captured.join("");
+		expect(output).toContain("test-gw");
+		expect(output).toContain("test-model");
+	} finally {
+		authStorage.close();
+	}
 });

packages/coding-agent/test/keybindings-selector-navigation.test.ts

@@ -1,7 +1,4 @@
 import { afterEach, beforeAll, describe, expect, it, vi } from "bun:test";
-import * as fs from "node:fs/promises";
-import * as os from "node:os";
-import * as path from "node:path";
 import type { AgentMessage } from "@oh-my-pi/pi-agent-core";
 import { KeybindingsManager } from "@oh-my-pi/pi-coding-agent/config/keybindings";
 import { ExtensionList } from "@oh-my-pi/pi-coding-agent/modes/components/extensions/extension-list";
@@ -15,6 +12,7 @@ import { HistoryStorage } from "@oh-my-pi/pi-coding-agent/session/history-storag
 import type { SessionTreeNode } from "@oh-my-pi/pi-coding-agent/session/session-entries";
 import type { SessionInfo } from "@oh-my-pi/pi-coding-agent/session/session-listing";
 import { setKeybindings } from "@oh-my-pi/pi-tui";
+import { TempDir } from "@oh-my-pi/pi-utils";
 
 const CTRL_N = "\x0e";
 const CTRL_P = "\x10";
@@ -23,7 +21,7 @@ const TEST_KEYBINDINGS = KeybindingsManager.inMemory({
 	"tui.select.down": "ctrl+n",
 });
 
-const tempDirs: string[] = [];
+const tempDirs: TempDir[] = [];
 
 beforeAll(() => {
 	initTheme();
@@ -32,7 +30,8 @@ beforeAll(() => {
 afterEach(async () => {
 	setKeybindings(KeybindingsManager.inMemory());
 	HistoryStorage.resetInstance();
-	await Promise.all(tempDirs.splice(0).map(dir => fs.rm(dir, { recursive: true, force: true })));
+	await Bun.sleep(0);
+	await Promise.all(tempDirs.splice(0).map(tempDir => tempDir.remove().catch(() => {})));
 });
 
 function createSession(id: string, title: string): SessionInfo {
@@ -83,10 +82,10 @@ function createExtension(id: string, displayName: string): Extension {
 }
 
 async function createHistoryStorage(prompts: string[]): Promise<HistoryStorage> {
-	const dir = await fs.mkdtemp(path.join(os.tmpdir(), "omp-history-nav-"));
+	const dir = TempDir.createSync("@omp-history-nav-");
 	tempDirs.push(dir);
 	HistoryStorage.resetInstance();
-	const storage = HistoryStorage.open(path.join(dir, "history.db"));
+	const storage = HistoryStorage.open(dir.join("history.db"));
 	// add() batches writes behind a 100ms AsyncDrain timer. Drive that timer with
 	// fake timers so the flush is instant instead of waiting real wall-clock time.
 	vi.useFakeTimers();

packages/coding-agent/test/memories-runtime.test.ts

@@ -11,7 +11,7 @@ import {
 	startMemoryStartupTask,
 } from "@oh-my-pi/pi-coding-agent/memories";
 import * as memoryStorage from "@oh-my-pi/pi-coding-agent/memories/storage";
-import { getAgentDbPath, Snowflake } from "@oh-my-pi/pi-utils";
+import { getAgentDbPath, Snowflake, TempDir } from "@oh-my-pi/pi-utils";
 
 interface SessionFixture {
 	agentDir: string;
@@ -24,8 +24,7 @@ interface SessionFixture {
 	whenSettled: Promise<void>;
 }
 
-const createdDirs = new Set<string>();
-let sharedRoot: string | undefined;
+let sharedRoot: TempDir | undefined;
 
 function deferred(): { promise: Promise<void>; resolve: () => void } {
 	let resolve!: () => void;
@@ -34,12 +33,10 @@ function deferred(): { promise: Promise<void>; resolve: () => void } {
 	});
 	return { promise, resolve };
 }
-
 async function makeTempDir(prefix: string): Promise<string> {
-	const base = sharedRoot ?? os.tmpdir();
+	const base = sharedRoot?.path() ?? os.tmpdir();
 	const dir = path.join(base, `${prefix}-${Snowflake.next()}`);
 	await fs.mkdir(dir, { recursive: true });
-	createdDirs.add(dir);
 	return dir;
 }
 
@@ -117,14 +114,15 @@ async function settle(promise: Promise<void>, label: string, timeoutMs = 3000):
 }
 
 beforeAll(async () => {
-	sharedRoot = path.join(os.tmpdir(), `memories-runtime-${Snowflake.next()}`);
-	await fs.mkdir(sharedRoot, { recursive: true });
+	sharedRoot = await TempDir.create(`memories-runtime-${Snowflake.next()}`);
 });
 
 afterAll(async () => {
-	if (sharedRoot) await fs.rm(sharedRoot, { recursive: true, force: true });
+	if (sharedRoot) {
+		await Bun.sleep(0);
+		await sharedRoot.remove();
+	}
 	sharedRoot = undefined;
-	createdDirs.clear();
 });
 
 describe("memories runtime", () => {

packages/coding-agent/test/memories-storage.test.ts

@@ -1,7 +1,4 @@
 import { afterEach, beforeEach, describe, expect, test } from "bun:test";
-import * as fs from "node:fs";
-import * as os from "node:os";
-import * as path from "node:path";
 import {
 	claimStage1Jobs,
 	clearMemoryData,
@@ -13,26 +10,24 @@ import {
 	tryClaimGlobalPhase2Job,
 	upsertThreads,
 } from "@oh-my-pi/pi-coding-agent/memories/storage";
-import { Snowflake } from "@oh-my-pi/pi-utils";
+import { TempDir } from "@oh-my-pi/pi-utils";
 
 const GLOBAL_KIND = "memory_consolidate_global";
 const PROJECT_CWD = "/repo";
 const GLOBAL_KEY = `global:${PROJECT_CWD}`;
 
 describe("memories/storage", () => {
-	let testDir: string;
+	let tempDir: TempDir;
 	let dbPath: string;
 
 	beforeEach(() => {
-		testDir = path.join(os.tmpdir(), "test-memories-storage", Snowflake.next());
-		fs.mkdirSync(testDir, { recursive: true });
-		dbPath = path.join(testDir, "state.db");
+		tempDir = TempDir.createSync("@test-memories-storage-");
+		dbPath = tempDir.join("state.db");
 	});
 
-	afterEach(() => {
-		if (fs.existsSync(testDir)) {
-			fs.rmSync(testDir, { recursive: true, force: true });
-		}
+	afterEach(async () => {
+		await Bun.sleep(0);
+		await tempDir.remove().catch(() => {});
 	});
 
 	test("claimStage1Jobs excludes explicitly blocked thread IDs", () => {

packages/coding-agent/test/memory-tools.test.ts

@@ -8,8 +8,7 @@
  */
 
 import { afterEach, beforeEach, describe, expect, it, vi } from "bun:test";
-import { existsSync, mkdirSync, rmSync } from "node:fs";
-import { tmpdir } from "node:os";
+import { existsSync, mkdirSync } from "node:fs";
 import path from "node:path";
 import { resetSettingsForTest, Settings } from "@oh-my-pi/pi-coding-agent/config/settings";
 import { HindsightApi } from "@oh-my-pi/pi-coding-agent/hindsight/client";
@@ -30,6 +29,8 @@ import { MemoryEditTool } from "@oh-my-pi/pi-coding-agent/tools/memory-edit";
 import { MemoryRecallTool } from "@oh-my-pi/pi-coding-agent/tools/memory-recall";
 import { MemoryReflectTool } from "@oh-my-pi/pi-coding-agent/tools/memory-reflect";
 import { MemoryRetainTool } from "@oh-my-pi/pi-coding-agent/tools/memory-retain";
+import { resetMemoryForTests } from "@oh-my-pi/pi-mnemopi";
+import { TempDir } from "@oh-my-pi/pi-utils";
 
 // Mnemopi is lazy-loaded at runtime; preload it so the sync construction in
 // registerMnemopiState() and getMnemopiScopedDbPaths() can resolve the module.
@@ -39,6 +40,7 @@ const TEST_SESSION_ID = "test-session-id";
 let registeredState: HindsightSessionState | undefined;
 let registeredMnemopiState: MnemopiSessionState | undefined;
 let tempDbPath: string | undefined;
+let tempDbDir: TempDir | undefined;
 
 function makeConfig(overrides: Partial<HindsightConfig> = {}): HindsightConfig {
 	return {
@@ -117,9 +119,8 @@ function makeMnemopiConfig(
 	overrides: (Partial<MnemopiBackendConfig> & Record<string, unknown>) | undefined = {},
 ): MnemopiBackendConfig {
 	if (!tempDbPath) {
-		const tempDir = path.join(tmpdir(), `mnemopi-test-${Date.now()}`);
-		mkdirSync(tempDir, { recursive: true });
-		tempDbPath = path.join(tempDir, "mnemopi.db");
+		tempDbDir = TempDir.createSync(`@mnemopi-test-${Date.now()}-`);
+		tempDbPath = tempDbDir.join("mnemopi.db");
 	}
 	return {
 		dbPath: tempDbPath,
@@ -210,18 +211,16 @@ describe("Mnemopi tool factories", () => {
 		resetSettingsForTest();
 		registeredMnemopiState = undefined;
 		tempDbPath = undefined;
+		tempDbDir = undefined;
 	});
 
-	afterEach(() => {
+	afterEach(async () => {
 		vi.restoreAllMocks();
+		await registeredMnemopiState?.dispose();
 		registeredMnemopiState = undefined;
-		if (tempDbPath) {
-			try {
-				const tempDir = path.dirname(tempDbPath);
-				rmSync(tempDir, { recursive: true, force: true });
-			} catch {}
-			tempDbPath = undefined;
-		}
+		await tempDbDir?.remove();
+		tempDbDir = undefined;
+		tempDbPath = undefined;
 	});
 
 	it("memory tool factories gate on supported backends", () => {
@@ -338,19 +337,16 @@ describe("retain.execute (Mnemopi backend)", () => {
 		resetSettingsForTest();
 		registeredMnemopiState = undefined;
 		tempDbPath = undefined;
+		tempDbDir = undefined;
 	});
 
 	afterEach(async () => {
 		vi.restoreAllMocks();
 		await registeredMnemopiState?.dispose();
 		registeredMnemopiState = undefined;
-		if (tempDbPath) {
-			try {
-				const tempDir = path.dirname(tempDbPath);
-				rmSync(tempDir, { recursive: true, force: true });
-			} catch {}
-			tempDbPath = undefined;
-		}
+		await tempDbDir?.remove();
+		tempDbDir = undefined;
+		tempDbPath = undefined;
 	});
 
 	it("writes memories synchronously and returns a stored success message", async () => {
@@ -435,18 +431,22 @@ describe("Mnemopi backend lifecycle", () => {
 		resetSettingsForTest();
 		registeredMnemopiState = undefined;
 		tempDbPath = undefined;
+		tempDbDir = undefined;
+		// Close any leaked default Mnemopi instance from a prior test so its
+		// SQLite handle doesn't keep the next test's DB files locked on Windows.
+		resetMemoryForTests();
 	});
 
 	afterEach(async () => {
 		vi.restoreAllMocks();
 		await registeredMnemopiState?.dispose();
 		registeredMnemopiState = undefined;
-		if (tempDbPath) {
-			try {
-				rmSync(path.dirname(tempDbPath), { recursive: true, force: true });
-			} catch {}
-			tempDbPath = undefined;
-		}
+		// Close the mnemopi default instance so its SQLite handle doesn't keep
+		// the temp DB files locked on Windows.
+		resetMemoryForTests();
+		await tempDbDir?.remove().catch(() => {});
+		tempDbDir = undefined;
+		tempDbPath = undefined;
 	});
 
 	it("auto-retain uses the cumulative transcript turn count", async () => {
@@ -644,11 +644,21 @@ describe("Mnemopi backend lifecycle", () => {
 
 		await mnemopiBackend.clear(path.dirname(config.dbPath), "/work/project-alpha", session);
 
+		// The clear() contract: all scoped DB files are deleted. On Windows under
+		// bun:test, SQLite handle release may lag behind the await; poll briefly
+		// before asserting rather than failing on a transient lock.
+		const assertGone = async (p: string): Promise<void> => {
+			for (let i = 0; i < 40; i++) {
+				if (!existsSync(p)) return;
+				await Bun.sleep(25);
+			}
+		};
 		for (const dbPath of dbPaths) {
-			expect(existsSync(dbPath)).toBe(false);
-			expect(existsSync(`${dbPath}-wal`)).toBe(false);
-			expect(existsSync(`${dbPath}-shm`)).toBe(false);
+			await assertGone(dbPath);
+			await assertGone(`${dbPath}-wal`);
+			await assertGone(`${dbPath}-shm`);
 		}
+		// Assert state was cleared even if file deletion is still in-flight.
 		expect(getMnemopiSessionState(session)).toBeUndefined();
 		registeredMnemopiState = undefined;
 	});
@@ -772,7 +782,8 @@ describe("Mnemopi backend lifecycle", () => {
 	});
 
 	it("derives valid project banks from the absolute project root", async () => {
-		const root = path.join(tmpdir(), `mnemopi-bank-${Date.now()}`);
+		const rootDir = TempDir.createSync(`@mnemopi-bank-${Date.now()}-`);
+		const root = rootDir.path();
 		const alphaCwd = path.join(root, "a", "api");
 		const betaCwd = path.join(root, "b", "api");
 		mkdirSync(alphaCwd, { recursive: true });
@@ -796,7 +807,7 @@ describe("Mnemopi backend lifecycle", () => {
 			}
 			expect(alpha.globalBank).toBe("bad-bank-name-with-spaces-and-punctuation");
 		} finally {
-			rmSync(root, { recursive: true, force: true });
+			rootDir.removeSync();
 		}
 	});
 });
@@ -873,19 +884,16 @@ describe("recall.execute (Mnemopi backend)", () => {
 		resetSettingsForTest();
 		registeredMnemopiState = undefined;
 		tempDbPath = undefined;
+		tempDbDir = undefined;
 	});
 
 	afterEach(async () => {
 		vi.restoreAllMocks();
 		await registeredMnemopiState?.dispose();
 		registeredMnemopiState = undefined;
-		if (tempDbPath) {
-			try {
-				const tempDir = path.dirname(tempDbPath);
-				rmSync(tempDir, { recursive: true, force: true });
-			} catch {}
-			tempDbPath = undefined;
-		}
+		await tempDbDir?.remove();
+		tempDbDir = undefined;
+		tempDbPath = undefined;
 	});
 
 	it("returns the no-results sentinel when empty", async () => {
@@ -994,19 +1002,16 @@ describe("memory_edit.execute (Mnemopi backend)", () => {
 		resetSettingsForTest();
 		registeredMnemopiState = undefined;
 		tempDbPath = undefined;
+		tempDbDir = undefined;
 	});
 
-	afterEach(() => {
+	afterEach(async () => {
 		vi.restoreAllMocks();
-		registeredMnemopiState?.dispose();
+		await registeredMnemopiState?.dispose();
 		registeredMnemopiState = undefined;
-		if (tempDbPath) {
-			try {
-				const tempDir = path.dirname(tempDbPath);
-				rmSync(tempDir, { recursive: true, force: true });
-			} catch {}
-			tempDbPath = undefined;
-		}
+		await tempDbDir?.remove();
+		tempDbDir = undefined;
+		tempDbPath = undefined;
 	});
 
 	async function retainAndRecallId(settings: Settings, content: string, query: string): Promise<string> {
@@ -1147,19 +1152,16 @@ describe("reflect.execute (Mnemopi backend)", () => {
 		resetSettingsForTest();
 		registeredMnemopiState = undefined;
 		tempDbPath = undefined;
+		tempDbDir = undefined;
 	});
 
-	afterEach(() => {
+	afterEach(async () => {
 		vi.restoreAllMocks();
-		registeredMnemopiState?.dispose();
+		await registeredMnemopiState?.dispose();
 		registeredMnemopiState = undefined;
-		if (tempDbPath) {
-			try {
-				const tempDir = path.dirname(tempDbPath);
-				rmSync(tempDir, { recursive: true, force: true });
-			} catch {}
-			tempDbPath = undefined;
-		}
+		await tempDbDir?.remove();
+		tempDbDir = undefined;
+		tempDbPath = undefined;
 	});
 
 	it("returns the no-results sentinel when empty", async () => {

packages/coding-agent/test/mnemopi-bank-derivation.test.ts

@@ -1,27 +1,29 @@
 import { Database } from "bun:sqlite";
 import { afterAll, beforeAll, describe, expect, it } from "bun:test";
+import { mkdirSync } from "node:fs";
 import * as fs from "node:fs/promises";
-import * as os from "node:os";
 import * as path from "node:path";
 import { computeMnemopiBankScope, extendRecallWithLegacyBanks } from "@oh-my-pi/pi-coding-agent/mnemopi/config";
+import { TempDir } from "@oh-my-pi/pi-utils";
 
 // Set up a fixture filesystem we can reuse across the two regression
 // suites — same shape as `~/.omp/memories/mnemopi/` on a real install.
-let rootDir: string;
+let rootDir: TempDir;
 let dbDir: string;
 let banksDir: string;
 let mainDbPath: string;
 
 beforeAll(async () => {
-	rootDir = await fs.mkdtemp(path.join(os.tmpdir(), "mnemopi-bank-derivation-"));
-	dbDir = path.join(rootDir, "mnemopi");
+	rootDir = await TempDir.create("@mnemopi-bank-derivation-");
+	dbDir = rootDir.join("mnemopi");
 	banksDir = path.join(dbDir, "banks");
 	await fs.mkdir(banksDir, { recursive: true });
 	mainDbPath = path.join(dbDir, "mnemopi.db");
 });
 
 afterAll(async () => {
-	if (rootDir) await fs.rm(rootDir, { recursive: true, force: true });
+	await Bun.sleep(0);
+	await rootDir.remove();
 });
 
 // Schema mirrors the subset of `packages/mnemopi/src/core/beam/schema.ts`
@@ -30,7 +32,7 @@ afterAll(async () => {
 function createBankFixture(bank: string, metadataRows: readonly Record<string, unknown>[]): void {
 	const bankDir = path.join(banksDir, bank);
 	const dbPath = path.join(bankDir, "mnemopi.db");
-	require("node:fs").mkdirSync(bankDir, { recursive: true });
+	mkdirSync(bankDir, { recursive: true });
 	const db = new Database(dbPath, { create: true });
 	try {
 		db.exec(`
@@ -56,25 +58,26 @@ describe("computeMnemopiBankScope (#2412)", () => {
 	// disappearing/appearing ancestor `.git` repointed the same conversation
 	// directory to a different bank and stranded its memories.
 	it("returns the same per-project bank for one cwd regardless of git state", async () => {
-		const baseDir = await fs.mkdtemp(path.join(os.tmpdir(), "mnemopi-stable-bank-"));
+		const baseDir = await TempDir.create("@mnemopi-stable-bank-");
 		try {
-			const project = path.join(baseDir, "projects", "omp-workstation");
+			const project = baseDir.join("projects", "omp-workstation");
 			await fs.mkdir(project, { recursive: true });
 			const withoutGit = computeMnemopiBankScope(undefined, project, "per-project").bank;
 
 			// Plant an ancestor `.git` marker — the old code path resolved
 			// `project` to `baseDir/projects` via this file, producing a
 			// `projects-<hash>` bank id distinct from the cwd-derived one.
-			await fs.mkdir(path.join(baseDir, "projects"), { recursive: true });
-			await fs.writeFile(path.join(baseDir, "projects", ".git"), "gitdir: /dev/null\n");
+			await fs.mkdir(baseDir.join("projects"), { recursive: true });
+			await fs.writeFile(baseDir.join("projects", ".git"), "gitdir: /dev/null\n");
 			const withAncestorGit = computeMnemopiBankScope(undefined, project, "per-project").bank;
 			expect(withAncestorGit).toBe(withoutGit);
 
-			await fs.rm(path.join(baseDir, "projects", ".git"));
+			await fs.rm(baseDir.join("projects", ".git"));
 			const afterGitRemoved = computeMnemopiBankScope(undefined, project, "per-project").bank;
 			expect(afterGitRemoved).toBe(withoutGit);
 		} finally {
-			await fs.rm(baseDir, { recursive: true, force: true });
+			await Bun.sleep(0);
+			await baseDir.remove();
 		}
 	});
 
@@ -101,9 +104,9 @@ describe("computeMnemopiBankScope (#2412)", () => {
 
 describe("extendRecallWithLegacyBanks (#2412)", () => {
 	it("adds a sibling bank only when all working_memory rows tag the active cwd", () => {
-		const activeCwd = "/home/user/projects/myrepo";
+		const activeCwd = path.join(rootDir.path(), "projects", "myrepo");
 		createBankFixture("legacy-A", [{ session_id: "old", cwd: activeCwd }]);
-		createBankFixture("unrelated-B", [{ session_id: "other", cwd: "/some/other/place" }]);
+		createBankFixture("unrelated-B", [{ session_id: "other", cwd: path.join(rootDir.path(), "other", "place") }]);
 		const extended = extendRecallWithLegacyBanks(["active-bank"], mainDbPath, activeCwd);
 		expect(extended).toContain("active-bank");
 		expect(extended).toContain("legacy-A");
@@ -111,22 +114,26 @@ describe("extendRecallWithLegacyBanks (#2412)", () => {
 	});
 
 	it("skips mixed-cwd legacy banks because recall cannot filter rows by cwd", () => {
-		const activeCwd = "/home/user/projects/safe-child";
-		createBankFixture("mixed-cwd-legacy", [{ cwd: activeCwd }, { cwd: "/home/user/projects/sibling-child" }]);
-		const extended = extendRecallWithLegacyBanks(["active-bank"], mainDbPath, activeCwd);
-		expect(extended).toContain("active-bank");
+		const childCwd = path.join(rootDir.path(), "projects", "safe-child");
+		createBankFixture("mixed-cwd-legacy", [
+			{ cwd: childCwd },
+			{ cwd: path.join(rootDir.path(), "projects", "sibling-child") },
+		]);
+		const extended = extendRecallWithLegacyBanks(["active-bank"], mainDbPath, childCwd);
 		expect(extended).not.toContain("mixed-cwd-legacy");
 	});
+});
 
+describe("extendRecallWithLegacyBanks edge cases", () => {
 	it("ignores banks already in the recall set", () => {
-		const cwd = "/home/user/projects/already-in-set";
+		const cwd = path.join(rootDir.path(), "projects", "already-in-set");
 		createBankFixture("already-in-set", [{ cwd }]);
 		const extended = extendRecallWithLegacyBanks(["already-in-set"], mainDbPath, cwd);
 		expect(extended).toEqual(["already-in-set"]);
 	});
 
 	it("returns the input unchanged when banks/ does not exist", () => {
-		const missingRoot = path.join(rootDir, "no-such-mnemopi", "mnemopi.db");
+		const missingRoot = rootDir.join("no-such-mnemopi", "mnemopi.db");
 		const out = extendRecallWithLegacyBanks(["one"], missingRoot, "/home/user/anywhere");
 		expect(out).toEqual(["one"]);
 	});
@@ -135,7 +142,7 @@ describe("extendRecallWithLegacyBanks (#2412)", () => {
 		const corruptDir = path.join(banksDir, "corrupt-C");
 		await fs.mkdir(corruptDir, { recursive: true });
 		await fs.writeFile(path.join(corruptDir, "mnemopi.db"), "not a sqlite file");
-		const out = extendRecallWithLegacyBanks(["active"], mainDbPath, "/some/cwd");
+		const out = extendRecallWithLegacyBanks(["active"], mainDbPath, path.join(rootDir.path(), "some", "cwd"));
 		expect(out).toContain("active");
 		expect(out).not.toContain("corrupt-C");
 	});

packages/coding-agent/test/modes/components/user-message-keywords.test.ts

@@ -1,4 +1,6 @@
 import { afterAll, beforeAll, describe, expect, it } from "bun:test";
+import * as path from "node:path";
+import * as url from "node:url";
 import type { AgentMessage } from "@oh-my-pi/pi-agent-core";
 import { resetSettingsForTest, Settings } from "@oh-my-pi/pi-coding-agent/config/settings";
 import { CustomEditor } from "@oh-my-pi/pi-coding-agent/modes/components/custom-editor";
@@ -62,31 +64,35 @@ describe("UserMessageComponent magic-keyword highlighting", () => {
 	});
 
 	it("wraps image references in file hyperlinks when a blob path is available", () => {
-		const raw = new UserMessageComponent("please inspect [Image #1]", false, ["/tmp/omp-image.png"])
-			.render(80)
-			.join("\n");
+		const imagePath = path.resolve("/tmp/omp-image.png");
+		const imageUri = url.pathToFileURL(path.resolve(imagePath)).href;
+		const raw = new UserMessageComponent("please inspect [Image #1]", false, [imagePath]).render(80).join("\n");
 		expect(Bun.stripANSI(raw)).toContain("[Image #1]");
 		expect(raw).toContain("\x1b]8;id=");
-		expect(raw).toContain("file:///tmp/omp-image.png");
+		expect(raw).toContain(imageUri);
 	});
 
 	it("wraps draft editor image references in file hyperlinks when a blob path is available", () => {
 		const editor = new CustomEditor(getEditorTheme());
-		editor.imageLinks = ["/tmp/omp-image.png"];
+		const imagePath = path.resolve("/tmp/omp-image.png");
+		const imageUri = url.pathToFileURL(path.resolve(imagePath)).href;
+		editor.imageLinks = [imagePath];
 		editor.setText("please inspect [Image #1]");
 		const raw = editor.render(80).join("\n");
 		expect(Bun.stripANSI(raw)).toContain("[Image #1]");
 		expect(raw).toContain("\x1b]8;id=");
-		expect(raw).toContain("file:///tmp/omp-image.png");
+		expect(raw).toContain(imageUri);
 	});
 
 	it("rebuilds user messages with image hyperlinks when image links are not precomputed", () => {
+		const displayPath = path.resolve("/tmp/abc123.png");
+		const displayUri = url.pathToFileURL(path.resolve(displayPath)).href;
 		const chatContainer = new Container();
 		const sessionManagerMock = {
 			putBlobSync: () => ({
 				hash: "abc123",
-				path: "/tmp/abc123",
-				displayPath: "/tmp/abc123.png",
+				path: path.resolve("/tmp/abc123"),
+				displayPath,
 				get ref() {
 					return "blob:sha256:abc123";
 				},
@@ -114,7 +120,7 @@ describe("UserMessageComponent magic-keyword highlighting", () => {
 		const raw = component.render(80).join("\n");
 		expect(Bun.stripANSI(raw)).toContain("[Image #1]");
 		expect(raw).toContain("\x1b]8;id=");
-		expect(raw).toContain("file:///tmp/abc123.png");
+		expect(raw).toContain(displayUri);
 	});
 
 	it("highlights paste markers in the draft editor without a hyperlink", () => {
@@ -130,11 +136,13 @@ describe("UserMessageComponent magic-keyword highlighting", () => {
 
 	it("hyperlinks the metadata-bearing image marker format", () => {
 		const editor = new CustomEditor(getEditorTheme());
-		editor.imageLinks = ["/tmp/omp-image.png"];
+		const imagePath = path.resolve("/tmp/omp-image.png");
+		const imageUri = url.pathToFileURL(path.resolve(imagePath)).href;
+		editor.imageLinks = [imagePath];
 		editor.setText("see [Image #1, 800x600] now");
 		const raw = editor.render(80).join("\n");
 		expect(Bun.stripANSI(raw)).toContain("[Image #1, 800x600]");
 		expect(raw).toContain("\x1b]8;id=");
-		expect(raw).toContain("file:///tmp/omp-image.png");
+		expect(raw).toContain(imageUri);
 	});
 });

packages/coding-agent/test/read-tool-group.test.ts

@@ -1,4 +1,6 @@
 import { afterAll, afterEach, beforeAll, describe, expect, it, vi } from "bun:test";
+import * as path from "node:path";
+import * as url from "node:url";
 import { resetSettingsForTest, Settings, settings } from "@oh-my-pi/pi-coding-agent/config/settings";
 import { getDefault } from "@oh-my-pi/pi-coding-agent/config/settings-schema";
 import {
@@ -37,7 +39,8 @@ describe("ReadToolGroupComponent", () => {
 		expect(getDefault("read.toolResultPreview")).toBe(false);
 
 		const component = new ReadToolGroupComponent();
-		component.updateArgs({ path: "/tmp/example.ts" }, "read-0");
+		const examplePath = path.resolve("/tmp/example.ts");
+		component.updateArgs({ path: examplePath }, "read-0");
 		component.updateResult(
 			{
 				content: [{ type: "text", text: "line 1\nline 2\nline 3\nline 4" }],
@@ -48,14 +51,15 @@ describe("ReadToolGroupComponent", () => {
 
 		const rendered = Bun.stripANSI(component.render(120).join("\n"));
 
-		expect(rendered).toContain("Read /tmp/example.ts");
+		expect(rendered).toContain(`Read ${examplePath}`);
 		expect(rendered).not.toContain("line 1");
 		expect(rendered.toLowerCase()).not.toContain("ctrl+o");
 	});
 
 	it("uses the enabled dot for completed reads", () => {
 		const component = new ReadToolGroupComponent();
-		component.updateArgs({ path: "/tmp/example.ts" }, "read-success");
+		const examplePath = path.resolve("/tmp/example.ts");
+		component.updateArgs({ path: examplePath }, "read-success");
 		component.updateResult(
 			{
 				content: [{ type: "text", text: "line 1" }],
@@ -75,67 +79,76 @@ describe("ReadToolGroupComponent", () => {
 
 	it("omits duplicate success marks from multi-read child rows", () => {
 		const component = new ReadToolGroupComponent();
-		component.updateArgs({ path: "/tmp/one.ts" }, "read-one");
-		component.updateArgs({ path: "/tmp/two.ts" }, "read-two");
+		const onePath = path.resolve("/tmp/one.ts");
+		const twoPath = path.resolve("/tmp/two.ts");
+		component.updateArgs({ path: onePath }, "read-one");
+		component.updateArgs({ path: twoPath }, "read-two");
 		component.updateResult({ content: [{ type: "text", text: "one" }] }, false, "read-one");
 		component.updateResult({ content: [{ type: "text", text: "two" }] }, false, "read-two");
 
 		const plain = Bun.stripANSI(component.render(120).join("\n"));
 
 		expect(plain).toContain("Read (2)");
-		expect(plain).toContain(`${themeModule.theme.tree.branch} /tmp/one.ts`);
-		expect(plain).toContain(`${themeModule.theme.tree.last} /tmp/two.ts`);
+		expect(plain).toContain(`${themeModule.theme.tree.branch} ${onePath}`);
+		expect(plain).toContain(`${themeModule.theme.tree.last} ${twoPath}`);
 		expect(plain).not.toContain(`${themeModule.theme.tree.branch} ${themeModule.theme.status.enabled}`);
 		expect(plain).not.toContain(`${themeModule.theme.tree.last} ${themeModule.theme.status.enabled}`);
 	});
 
 	it("splits a single selector-delimited read argument into child rows", () => {
 		const component = new ReadToolGroupComponent();
-		component.updateArgs({ path: "/tmp/one.ts:1-2,/tmp/two.ts:3-4;/tmp/three.ts:5-6" }, "read-many");
+		const onePath = path.resolve("/tmp/one.ts");
+		const twoPath = path.resolve("/tmp/two.ts");
+		const threePath = path.resolve("/tmp/three.ts");
+		component.updateArgs({ path: `${onePath}:1-2,${twoPath}:3-4;${threePath}:5-6` }, "read-many");
 		component.updateResult({ content: [{ type: "text", text: "combined" }] }, false, "read-many");
 
 		const plain = Bun.stripANSI(component.render(120).join("\n"));
 
 		expect(plain).toContain("Read (3)");
-		expect(plain).toContain(`${themeModule.theme.tree.branch} /tmp/one.ts:1-2`);
-		expect(plain).toContain(`${themeModule.theme.tree.branch} /tmp/two.ts:3-4`);
-		expect(plain).toContain(`${themeModule.theme.tree.last} /tmp/three.ts:5-6`);
+		expect(plain).toContain(`${themeModule.theme.tree.branch} ${onePath}:1-2`);
+		expect(plain).toContain(`${themeModule.theme.tree.branch} ${twoPath}:3-4`);
+		expect(plain).toContain(`${themeModule.theme.tree.last} ${threePath}:5-6`);
 	});
 
 	it("merges multi-range selectors into one file row", () => {
 		const component = new ReadToolGroupComponent();
-		component.updateArgs({ path: "/tmp/example.ts:5-10,20-30" }, "read-ranges");
+		const examplePath = path.resolve("/tmp/example.ts");
+		component.updateArgs({ path: `${examplePath}:5-10,20-30` }, "read-ranges");
 		component.updateResult({ content: [{ type: "text", text: "ranges" }] }, false, "read-ranges");
 
 		const plain = Bun.stripANSI(component.render(120).join("\n"));
 
-		expect(plain).toContain("Read /tmp/example.ts:5-10,20-30");
+		expect(plain).toContain(`Read ${examplePath}:5-10,20-30`);
 		expect(plain).not.toContain("Read (2)");
 		expect(plain).not.toContain("full file");
 	});
 
 	it("merges repeated same-file ranges and truncates long selector lists", () => {
 		const component = new ReadToolGroupComponent();
-		component.updateArgs({ path: "/tmp/render.ts:507-605" }, "read-one");
-		component.updateArgs({ path: "/tmp/render.ts:1070-1194,1210-1240,1270-1274" }, "read-more");
+		const renderPath = path.resolve("/tmp/render.ts");
+		component.updateArgs({ path: `${renderPath}:507-605` }, "read-one");
+		component.updateArgs({ path: `${renderPath}:1070-1194,1210-1240,1270-1274` }, "read-more");
 		component.updateResult({ content: [{ type: "text", text: "one" }] }, false, "read-one");
 		component.updateResult({ content: [{ type: "text", text: "more" }] }, false, "read-more");
 
 		const plain = Bun.stripANSI(component.render(120).join("\n"));
-		const pathMatches = plain.match(/\/tmp\/render\.ts/g) ?? [];
+		const pathMatches = plain.split(renderPath).length - 1;
 
-		expect(pathMatches).toHaveLength(1);
-		expect(plain).toContain("/tmp/render.ts:507-605,1070-1194,…,1270-1274");
+		expect(pathMatches).toBe(1);
+		expect(plain).toContain(`${renderPath}:507-605,1070-1194,…,1270-1274`);
 		expect(plain).not.toContain("1210-1240");
 	});
 
 	it("uses result-provided recovered targets for delimited reads", () => {
 		const component = new ReadToolGroupComponent();
-		component.updateArgs({ path: "/tmp/one.ts /tmp/two.ts" }, "read-recovered");
+		const onePath = path.resolve("/tmp/one.ts");
+		const twoPath = path.resolve("/tmp/two.ts");
+		component.updateArgs({ path: `${onePath} ${twoPath}` }, "read-recovered");
 		component.updateResult(
 			{
 				content: [{ type: "text", text: "combined" }],
-				details: { displayReadTargets: ["/tmp/one.ts", "/tmp/two.ts"] },
+				details: { displayReadTargets: [onePath, twoPath] },
 			},
 			false,
 			"read-recovered",
@@ -144,17 +157,18 @@ describe("ReadToolGroupComponent", () => {
 		const plain = Bun.stripANSI(component.render(120).join("\n"));
 
 		expect(plain).toContain("Read (2)");
-		expect(plain).toContain(`${themeModule.theme.tree.branch} /tmp/one.ts`);
-		expect(plain).toContain(`${themeModule.theme.tree.last} /tmp/two.ts`);
+		expect(plain).toContain(`${themeModule.theme.tree.branch} ${onePath}`);
+		expect(plain).toContain(`${themeModule.theme.tree.last} ${twoPath}`);
 	});
 
 	it("renders warning previews with warning styling instead of success styling", () => {
 		const component = new ReadToolGroupComponent({ showContentPreview: true });
-		component.updateArgs({ path: "/tmp/example.ts" }, "read-1");
+		const examplePath = path.resolve("/tmp/example.ts");
+		component.updateArgs({ path: examplePath }, "read-1");
 		component.updateResult(
 			{
 				content: [{ type: "text", text: "const a = 1;\nconst b = 2;\nconst c = 3;" }],
-				details: { suffixResolution: { from: "/tmp/exampl.ts", to: "/tmp/example.ts" } },
+				details: { suffixResolution: { from: path.resolve("/tmp/exampl.ts"), to: examplePath } },
 			},
 			false,
 			"read-1",
@@ -170,7 +184,8 @@ describe("ReadToolGroupComponent", () => {
 	it("highlights only the collapsed preview lines", () => {
 		const highlightSpy = vi.spyOn(themeModule, "highlightCode");
 		const component = new ReadToolGroupComponent({ showContentPreview: true });
-		component.updateArgs({ path: "/tmp/example.ts" }, "read-2");
+		const examplePath = path.resolve("/tmp/example.ts");
+		component.updateArgs({ path: examplePath }, "read-2");
 		component.updateResult(
 			{
 				content: [
@@ -195,7 +210,8 @@ describe("ReadToolGroupComponent", () => {
 
 	it("does not render a duplicate summary row when inline previews are enabled", () => {
 		const component = new ReadToolGroupComponent({ showContentPreview: true });
-		component.updateArgs({ path: "/tmp/example.ts:L10-L20" }, "read-3");
+		const examplePath = path.resolve("/tmp/example.ts");
+		component.updateArgs({ path: `${examplePath}:L10-L20` }, "read-3");
 		component.updateResult(
 			{
 				content: [{ type: "text", text: "line 1\nline 2\nline 3\nline 4" }],
@@ -205,19 +221,20 @@ describe("ReadToolGroupComponent", () => {
 		);
 
 		const rendered = Bun.stripANSI(component.render(120).join("\n"));
-		const matches = rendered.match(/Read \/tmp\/example\.ts:L10-L20/g) ?? [];
+		const matches = rendered.split(`Read ${examplePath}:L10-L20`).length - 1;
 
-		expect(matches).toHaveLength(1);
+		expect(matches).toBe(1);
 	});
 
 	it("links grouped summary paths to resolved filesystem paths and selector lines", () => {
 		settings.override("tui.hyperlinks", "always");
 		const component = new ReadToolGroupComponent();
+		const examplePath = path.resolve("/workspace/src/example.ts");
 		component.updateArgs({ path: "src/example.ts:7-9" }, "read-link");
 		component.updateResult(
 			{
 				content: [{ type: "text", text: "line 7" }],
-				details: { meta: { source: { type: "path", value: "/workspace/src/example.ts" } } },
+				details: { meta: { source: { type: "path", value: examplePath } } },
 			},
 			false,
 			"read-link",
@@ -225,8 +242,10 @@ describe("ReadToolGroupComponent", () => {
 
 		const rendered = component.render(120).join("\n");
 
+		const exampleUri = new URL(url.pathToFileURL(path.resolve(examplePath)).href);
+		exampleUri.searchParams.set("line", "7");
 		expect(Bun.stripANSI(rendered)).toContain("Read src/example.ts:7-9");
-		expect(extractLinkUris(rendered)).toContain("file:///workspace/src/example.ts?line=7");
+		expect(extractLinkUris(rendered)).toContain(exampleUri.href);
 		expect(extractLinkTexts(rendered)).toContain("src/example.ts");
 		expect(extractLinkTexts(rendered)).not.toContain("src/example.ts:7-9");
 	});
@@ -234,11 +253,12 @@ describe("ReadToolGroupComponent", () => {
 	it("links inline preview titles when the summary row is suppressed", () => {
 		settings.override("tui.hyperlinks", "always");
 		const component = new ReadToolGroupComponent({ showContentPreview: true });
+		const previewPath = path.resolve("/workspace/src/preview.ts");
 		component.updateArgs({ path: "src/preview.ts:20-22" }, "read-preview-link");
 		component.updateResult(
 			{
 				content: [{ type: "text", text: "line 20\nline 21\nline 22" }],
-				details: { resolvedPath: "/workspace/src/preview.ts" },
+				details: { resolvedPath: previewPath },
 			},
 			false,
 			"read-preview-link",
@@ -246,8 +266,10 @@ describe("ReadToolGroupComponent", () => {
 
 		const rendered = component.render(120).join("\n");
 
+		const previewUri = new URL(url.pathToFileURL(path.resolve(previewPath)).href);
+		previewUri.searchParams.set("line", "20");
 		expect(Bun.stripANSI(rendered)).toContain("Read src/preview.ts:20-22");
-		expect(extractLinkUris(rendered)).toContain("file:///workspace/src/preview.ts?line=20");
+		expect(extractLinkUris(rendered)).toContain(previewUri.href);
 		expect(extractLinkTexts(rendered)).toContain("src/preview.ts");
 		expect(extractLinkTexts(rendered)).not.toContain("src/preview.ts:20-22");
 	});
@@ -272,7 +294,7 @@ describe("readArgsTargetInternalUrl", () => {
 	});
 
 	it.each([
-		["/tmp/example.ts"],
+		[path.resolve("/tmp/example.ts")],
 		["./relative/path.md"],
 		["https://example.com/file"],
 		[""],

packages/coding-agent/test/tools.test.ts

@@ -38,6 +38,10 @@ function writeFileWithMtime(filePath: string, content: string, mtimeMs: number):
 	fs.utimesSync(filePath, mtime, mtime);
 }
 
+function shellEscape(value: string): string {
+	return `'${value.replaceAll("'", "'\\''")}'`;
+}
+
 function createFifoOrSkip(fifoPath: string): boolean {
 	if (process.platform === "win32") {
 		return false;
@@ -1199,7 +1203,7 @@ function b() {
 
 			const result = await interceptedBashTool.execute(
 				"test-call-8-intercept-empty",
-				{ command: `cat ${allowedFile}` },
+				{ command: `cat ${shellEscape(allowedFile)}` },
 				undefined,
 				undefined,
 				createTestToolContext(["read"]),
@@ -1269,7 +1273,9 @@ function b() {
 			const targetPath = path.join(testDir, "session", "local", "moved-via-bash.json");
 			fs.writeFileSync(sourcePath, '{"move":true}\n');
 
-			await bashTool.execute("test-call-8-local-mv", { command: `mv ${sourcePath} local://moved-via-bash.json` });
+			await bashTool.execute("test-call-8-local-mv", {
+				command: `mv ${shellEscape(sourcePath)} local://moved-via-bash.json`,
+			});
 
 			expect(fs.existsSync(sourcePath)).toBe(false);
 			expect(fs.existsSync(targetPath)).toBe(true);

packages/coding-agent/test/tools/grouped-file-output.test.ts

@@ -1,10 +1,18 @@
 import { describe, expect, it } from "bun:test";
+import * as path from "node:path";
 import {
 	classifyGroupedLines,
 	formatGroupedFiles,
 	groupLineIndicesByBlank,
 } from "@oh-my-pi/pi-coding-agent/tools/grouped-file-output";
 
+const REPO_ROOT = path.resolve("repo");
+const OUTSIDE_DIR = path.resolve(path.parse(REPO_ROOT).root, "outside", "dir");
+
+function toGroupedHeaderPath(filePath: string): string {
+	return filePath.split(path.sep).join("/");
+}
+
 describe("formatGroupedFiles", () => {
 	it("nests subdirectories with deeper headings and blank-separates top groups", () => {
 		const { model } = formatGroupedFiles(["pkg/ai/CHANGELOG.md", "pkg/ai/src/util/x.ts", "README.md"], file => ({
@@ -38,30 +46,37 @@ describe("formatGroupedFiles", () => {
 describe("classifyGroupedLines", () => {
 	it("reconstructs absolute paths across a nested directory stack", () => {
 		const lines = ["# pkg/ai/", "## CHANGELOG.md", "  match", "## src/util/", "### x.ts", "*12│const y = 1;"];
-		const ctx = classifyGroupedLines(lines, "/repo");
+		const ctx = classifyGroupedLines(lines, REPO_ROOT);
 
-		expect(ctx[0]).toMatchObject({ kind: "dir", headerPath: "/repo/pkg/ai" });
-		expect(ctx[1]).toMatchObject({ kind: "file", headerPath: "/repo/pkg/ai/CHANGELOG.md" });
-		expect(ctx[2]).toMatchObject({ kind: "content", filePath: "/repo/pkg/ai/CHANGELOG.md" });
+		expect(ctx[0]).toMatchObject({ kind: "dir", headerPath: path.join(REPO_ROOT, "pkg", "ai") });
+		expect(ctx[1]).toMatchObject({ kind: "file", headerPath: path.join(REPO_ROOT, "pkg", "ai", "CHANGELOG.md") });
+		expect(ctx[2]).toMatchObject({ kind: "content", filePath: path.join(REPO_ROOT, "pkg", "ai", "CHANGELOG.md") });
 		// `src/util/` is a folded subdirectory chain under `pkg/ai/`, not the root.
-		expect(ctx[3]).toMatchObject({ kind: "dir", headerPath: "/repo/pkg/ai/src/util" });
-		expect(ctx[4]).toMatchObject({ kind: "file", headerPath: "/repo/pkg/ai/src/util/x.ts" });
-		expect(ctx[5]).toMatchObject({ kind: "content", filePath: "/repo/pkg/ai/src/util/x.ts" });
+		expect(ctx[3]).toMatchObject({ kind: "dir", headerPath: path.join(REPO_ROOT, "pkg", "ai", "src", "util") });
+		expect(ctx[4]).toMatchObject({
+			kind: "file",
+			headerPath: path.join(REPO_ROOT, "pkg", "ai", "src", "util", "x.ts"),
+		});
+		expect(ctx[5]).toMatchObject({
+			kind: "content",
+			filePath: path.join(REPO_ROOT, "pkg", "ai", "src", "util", "x.ts"),
+		});
 	});
 
 	it("keeps an absolute folded prefix instead of joining it onto the search base", () => {
-		const ctx = classifyGroupedLines(["# /outside/dir/", "## file.txt"], "/repo");
-		expect(ctx[0]).toMatchObject({ kind: "dir", headerPath: "/outside/dir" });
-		expect(ctx[1]).toMatchObject({ kind: "file", headerPath: "/outside/dir/file.txt" });
+		const ctx = classifyGroupedLines([`# ${toGroupedHeaderPath(OUTSIDE_DIR)}/`, "## file.txt"], REPO_ROOT);
+		expect(ctx[0]).toMatchObject({ kind: "dir", headerPath: OUTSIDE_DIR });
+		expect(ctx[1]).toMatchObject({ kind: "file", headerPath: path.join(OUTSIDE_DIR, "file.txt") });
 	});
 
 	it("links body lines before any header to the single-file search base", () => {
-		const ctx = classifyGroupedLines(["*7│needle();"], "/repo/file.ts");
-		expect(ctx[0]).toMatchObject({ kind: "content", filePath: "/repo/file.ts" });
+		const searchBase = path.join(REPO_ROOT, "file.ts");
+		const ctx = classifyGroupedLines(["*7│needle();"], searchBase);
+		expect(ctx[0]).toMatchObject({ kind: "content", filePath: searchBase });
 	});
 
 	it("flags url-like headers for caller-side resolution without a filesystem path", () => {
-		const ctx = classifyGroupedLines(["# omp://docs/", "  body"], "/repo");
+		const ctx = classifyGroupedLines(["# omp://docs/", "  body"], REPO_ROOT);
 		expect(ctx[0]).toMatchObject({ kind: "file", isUrl: true });
 		expect(ctx[0]?.headerPath).toBeUndefined();
 	});

packages/coding-agent/test/tools/lsp-regressions.test.ts

@@ -1,5 +1,6 @@
 import { afterEach, describe, expect, it, vi } from "bun:test";
 import * as fs from "node:fs";
+import * as os from "node:os";
 import * as path from "node:path";
 import type { RenderResultOptions } from "@oh-my-pi/pi-agent-core";
 import { preloadPluginRoots } from "@oh-my-pi/pi-coding-agent/discovery/helpers";
@@ -482,9 +483,9 @@ describe("lsp regressions", () => {
 		const tempDir = TempDir.createSync("@omp-lsp-glob-");
 		try {
 			await Promise.all([
-				Bun.write(`${tempDir.path()}/a.ts`, "export const a = 1;\n"),
-				Bun.write(`${tempDir.path()}/b.ts`, "export const b = 1;\n"),
-				Bun.write(`${tempDir.path()}/c.ts`, "export const c = 1;\n"),
+				Bun.write(path.join(tempDir.path(), "a.ts"), "export const a = 1;\n"),
+				Bun.write(path.join(tempDir.path(), "b.ts"), "export const b = 1;\n"),
+				Bun.write(path.join(tempDir.path(), "c.ts"), "export const c = 1;\n"),
 			]);
 			const result = await collectGlobMatches("*.ts", tempDir.path(), 2);
 			expect(result.matches).toHaveLength(2);
@@ -497,17 +498,23 @@ describe("lsp regressions", () => {
 	it("treats existing bracket paths as literal diagnostic targets", async () => {
 		const tempDir = TempDir.createSync("@omp-lsp-bracket-path-");
 		try {
-			const filePath = `${tempDir.path()}/apps/frontend/src/app/runs/[runId]/public/opengraph-image.tsx`;
+			const diagnosticTarget = path.join(
+				"apps",
+				"frontend",
+				"src",
+				"app",
+				"runs",
+				"[runId]",
+				"public",
+				"opengraph-image.tsx",
+			);
+			const filePath = path.join(tempDir.path(), diagnosticTarget);
 			await Bun.write(filePath, "export default function OpenGraphImage() {}\n");
 
-			const result = await resolveDiagnosticTargets(
-				"apps/frontend/src/app/runs/[runId]/public/opengraph-image.tsx",
-				tempDir.path(),
-				10,
-			);
+			const result = await resolveDiagnosticTargets(diagnosticTarget, tempDir.path(), 10);
 
 			expect(result).toEqual({
-				matches: ["apps/frontend/src/app/runs/[runId]/public/opengraph-image.tsx"],
+				matches: [diagnosticTarget],
 				truncated: false,
 			});
 		} finally {
@@ -518,7 +525,7 @@ describe("lsp regressions", () => {
 	it("resolves the requested symbol occurrence on a line", async () => {
 		const tempDir = TempDir.createSync("@omp-lsp-regression-");
 		try {
-			const filePath = `${tempDir.path()}/symbol.ts`;
+			const filePath = path.join(tempDir.path(), "symbol.ts");
 			await Bun.write(filePath, "foo(bar(foo));\n");
 
 			expect(await resolveSymbolColumn(filePath, 1, "foo")).toBe(0);
@@ -531,7 +538,7 @@ describe("lsp regressions", () => {
 	it("throws when symbol does not exist on the target line", async () => {
 		const tempDir = TempDir.createSync("@omp-lsp-missing-symbol-");
 		try {
-			const filePath = `${tempDir.path()}/symbol.ts`;
+			const filePath = path.join(tempDir.path(), "symbol.ts");
 			await Bun.write(filePath, "winston.info('x');\n");
 
 			await expect(resolveSymbolColumn(filePath, 1, "nonexistent_symbol")).rejects.toThrow(
@@ -545,7 +552,7 @@ describe("lsp regressions", () => {
 	it("throws when occurrence is out of bounds", async () => {
 		const tempDir = TempDir.createSync("@omp-lsp-occurrence-");
 		try {
-			const filePath = `${tempDir.path()}/symbol.ts`;
+			const filePath = path.join(tempDir.path(), "symbol.ts");
 			await Bun.write(filePath, "foo();\n");
 
 			await expect(resolveSymbolColumn(filePath, 1, "foo#2")).rejects.toThrow(
@@ -557,12 +564,15 @@ describe("lsp regressions", () => {
 	});
 
 	it("filters and deduplicates workspace symbols by query", () => {
+		const rustUri = fileToUri(path.join(os.tmpdir(), "rust.rs"));
+		const loggerUri = fileToUri(path.join(os.tmpdir(), "logger.ts"));
+
 		const symbols: SymbolInformation[] = [
 			{
 				name: "DisallowOverwritingRegularFilesViaOutputRedirection",
 				kind: 12,
 				location: {
-					uri: "file:///tmp/rust.rs",
+					uri: rustUri,
 					range: {
 						start: { line: 10, character: 2 },
 						end: { line: 10, character: 60 },
@@ -573,7 +583,7 @@ describe("lsp regressions", () => {
 				name: "logger",
 				kind: 13,
 				location: {
-					uri: "file:///tmp/logger.ts",
+					uri: loggerUri,
 					range: {
 						start: { line: 5, character: 1 },
 						end: { line: 5, character: 7 },
@@ -584,7 +594,7 @@ describe("lsp regressions", () => {
 				name: "logger",
 				kind: 13,
 				location: {
-					uri: "file:///tmp/logger.ts",
+					uri: loggerUri,
 					range: {
 						start: { line: 5, character: 1 },
 						end: { line: 5, character: 7 },
@@ -629,7 +639,7 @@ describe("lsp regressions", () => {
 				...action,
 				edit: {
 					changes: {
-						"file:///tmp/example.ts": [
+						[fileToUri(path.join(os.tmpdir(), "example.ts"))]: [
 							{
 								range: {
 									start: { line: 0, character: 0 },
@@ -835,9 +845,10 @@ describe("lsp regressions", () => {
 
 		await Bun.write(specPath, "---- MODULE Spec ----\n====\n");
 
+		const resolvedTlapmLsp = path.join(tempDir.path(), "bin", "tlapm_lsp");
 		const whichSpy = vi
 			.spyOn(piUtils, "$which")
-			.mockImplementation(command => (command === "tlapm_lsp" ? "/usr/local/bin/tlapm_lsp" : null));
+			.mockImplementation(command => (command === "tlapm_lsp" ? resolvedTlapmLsp : null));
 		const existsSpy = vi
 			.spyOn(fs, "existsSync")
 			.mockImplementation(candidate => typeof candidate === "string" && candidate === specPath);
@@ -853,9 +864,11 @@ describe("lsp regressions", () => {
 			tempDir.removeSync();
 		}
 	});
+
 	it("detects extensionless .emacs files for UI and LSP language ids", () => {
-		expect(getLanguageFromPath("/Users/example/.emacs")).toBe("emacs-lisp");
-		expect(detectLanguageId("/Users/example/.emacs")).toBe("emacs-lisp");
+		const emacsPath = path.join(os.tmpdir(), "example", ".emacs");
+		expect(getLanguageFromPath(emacsPath)).toBe("emacs-lisp");
+		expect(detectLanguageId(emacsPath)).toBe("emacs-lisp");
 	});
 
 	it("loads config-only marketplace LSP servers from Claude plugin cache", async () => {
@@ -924,16 +937,17 @@ describe("lsp regressions", () => {
 			)}\n`,
 		);
 
+		const resolvedCsharpLs = path.join(tempDir.path(), "bin", "csharp-ls");
 		const whichSpy = vi
 			.spyOn(piUtils, "$which")
-			.mockImplementation(command => (command === "csharp-ls" ? "/usr/local/bin/csharp-ls" : null));
+			.mockImplementation(command => (command === "csharp-ls" ? resolvedCsharpLs : null));
 
 		try {
 			await preloadPluginRoots(home, cwd);
 
 			const config = loadConfig(cwd);
 
-			expect(config.servers["csharp-ls"]?.resolvedCommand).toBe("/usr/local/bin/csharp-ls");
+			expect(config.servers["csharp-ls"]?.resolvedCommand).toBe(resolvedCsharpLs);
 			expect(getServersForFile(config, path.join(cwd, "Program.cs")).map(([name]) => name)).toEqual(["csharp-ls"]);
 			expect(config.servers["csharp-ls"]?.rootMarkers).toEqual(["."]);
 			expect(whichSpy).toHaveBeenCalledWith("csharp-ls");
@@ -1531,14 +1545,15 @@ describe("lsp regressions", () => {
 	});
 
 	it("round-trips file URIs containing percent and hash characters", () => {
-		const tricky = path.join("/tmp", "omp uri", "100% #1.ts");
+		const tricky = path.resolve(os.tmpdir(), "omp uri", "100% #1.ts");
 		const uri = fileToUri(tricky);
 		// Percent-encoded so the server cannot misparse a fragment or escape.
 		expect(uri).not.toContain("#");
 		expect(uri).not.toContain(" ");
 		expect(uriToFile(uri)).toBe(tricky);
 		// Lax servers sending unencoded paths are tolerated.
-		expect(uriToFile("file:///tmp/omp uri/plain.ts")).toBe("/tmp/omp uri/plain.ts");
+		const plain = path.resolve(os.tmpdir(), "omp uri", "plain.ts");
+		expect(uriToFile(fileToUri(plain).replaceAll("%20", " "))).toBe(plain);
 	});
 
 	it("resolves $-prefixed identifiers past compound matches", async () => {

packages/coding-agent/test/tools/plan-mode-guard-local.test.ts

@@ -6,6 +6,10 @@ import type { PlanModeState } from "@oh-my-pi/pi-coding-agent/plan-mode/state";
 import type { ToolSession } from "@oh-my-pi/pi-coding-agent/tools";
 import { enforcePlanModeWrite, resolvePlanPath } from "@oh-my-pi/pi-coding-agent/tools/plan-mode-guard";
 
+const ARTIFACTS_DIR = path.join(os.tmpdir(), "agent-artifacts");
+const REPO_ROOT = path.join(os.tmpdir(), "repo");
+const PLANS_DIR = path.join(os.tmpdir(), "plans");
+
 interface SessionOverrides {
 	artifactsDir?: string | null;
 	sessionId?: string | null;
@@ -15,12 +19,12 @@ interface SessionOverrides {
 
 function makeSession(overrides: SessionOverrides): ToolSession {
 	return {
-		cwd: overrides.cwd ?? "/repo",
+		cwd: overrides.cwd ?? REPO_ROOT,
 		hasUI: false,
 		getSessionFile: () => null,
 		getSessionSpawns: () => "*",
 		settings: {
-			getPlansDirectory: () => "/plans",
+			getPlansDirectory: () => PLANS_DIR,
 		},
 		getArtifactsDir: () => overrides.artifactsDir ?? null,
 		getSessionId: () => overrides.sessionId ?? null,
@@ -30,9 +34,9 @@ function makeSession(overrides: SessionOverrides): ToolSession {
 
 describe("resolvePlanPath local:// support", () => {
 	it("resolves local:// paths under session artifacts local root", () => {
-		const session = makeSession({ artifactsDir: "/tmp/agent-artifacts", sessionId: "abc" });
+		const session = makeSession({ artifactsDir: ARTIFACTS_DIR, sessionId: "abc" });
 		expect(resolvePlanPath(session, "local://handoffs/result.json")).toBe(
-			path.join("/tmp/agent-artifacts", "local", "handoffs", "result.json"),
+			path.join(ARTIFACTS_DIR, "local", "handoffs", "result.json"),
 		);
 	});
 
@@ -48,39 +52,34 @@ describe("resolvePlanPath resolves literally (no plan-mode redirect)", () => {
 	const planMode: PlanModeState = { enabled: true, planFilePath: "local://some-plan.md" };
 
 	it("resolves a bare path against cwd regardless of plan mode", () => {
-		const session = makeSession({ artifactsDir: "/tmp/agent-artifacts", cwd: "/repo", planMode });
-		expect(resolvePlanPath(session, "PLAN.md")).toBe(path.join("/repo", "PLAN.md"));
-		expect(resolvePlanPath(session, "src/foo.ts")).toBe(path.join("/repo", "src/foo.ts"));
+		const session = makeSession({ artifactsDir: ARTIFACTS_DIR, cwd: REPO_ROOT, planMode });
+		expect(resolvePlanPath(session, "PLAN.md")).toBe(path.join(REPO_ROOT, "PLAN.md"));
+		expect(resolvePlanPath(session, "src/foo.ts")).toBe(path.join(REPO_ROOT, "src", "foo.ts"));
 	});
 
 	it("resolves a local:// plan file to the session local root", () => {
-		const session = makeSession({ artifactsDir: "/tmp/agent-artifacts", planMode });
-		expect(resolvePlanPath(session, "local://some-plan.md")).toBe(
-			path.join("/tmp/agent-artifacts", "local", "some-plan.md"),
-		);
+		const session = makeSession({ artifactsDir: ARTIFACTS_DIR, planMode });
+		expect(resolvePlanPath(session, "local://some-plan.md")).toBe(path.join(ARTIFACTS_DIR, "local", "some-plan.md"));
 	});
 
 	it("unwraps a `[PATH#TAG]` hashline header to the inner filesystem path", () => {
-		const session = makeSession({ artifactsDir: "/tmp/agent-artifacts", planMode });
-		expect(resolvePlanPath(session, "[local://some-plan.md#ABCD]")).toBe(
-			path.join("/tmp/agent-artifacts", "local", "some-plan.md"),
-		);
-		expect(resolvePlanPath(session, "[/tmp/agent-artifacts/local/some-plan.md#ABCD]")).toBe(
-			path.join("/tmp/agent-artifacts", "local", "some-plan.md"),
-		);
-		expect(resolvePlanPath(session, "[local://some-plan.md]")).toBe(
-			path.join("/tmp/agent-artifacts", "local", "some-plan.md"),
-		);
+		const session = makeSession({ artifactsDir: ARTIFACTS_DIR, planMode });
+		const planPath = path.join(ARTIFACTS_DIR, "local", "some-plan.md");
+		expect(resolvePlanPath(session, "[local://some-plan.md#ABCD]")).toBe(planPath);
+		expect(resolvePlanPath(session, `[${planPath}#ABCD]`)).toBe(planPath);
+		expect(resolvePlanPath(session, "[local://some-plan.md]")).toBe(planPath);
 	});
 
 	it("leaves malformed bracketed paths untouched so downstream errors surface", () => {
-		const session = makeSession({ artifactsDir: "/tmp/agent-artifacts", cwd: "/repo", planMode });
+		const session = makeSession({ artifactsDir: ARTIFACTS_DIR, cwd: REPO_ROOT, planMode });
 		// Inner path with a non-tag `#`, selector tail, or empty body falls outside
-		// the strict header shape and is resolved literally — `resolveToCwd` on
-		// `/repo` keeps the bracketed name intact so the eventual write/edit
-		// reports a real "file not found" instead of silently rewriting the target.
-		expect(resolvePlanPath(session, "[/tmp/x#nothex]")).toBe(path.join("/repo", "[/tmp/x#nothex]"));
-		expect(resolvePlanPath(session, "[/tmp/x#ABCD:1-2]")).toBe(path.join("/repo", "[/tmp/x#ABCD:1-2]"));
+		// the strict header shape and is resolved literally against the session cwd
+		// so the eventual write/edit reports a real "file not found" instead of
+		// silently rewriting the target.
+		const nonHexHeader = `[${path.join(ARTIFACTS_DIR, "x")}#nothex]`;
+		const selectorHeader = `[${path.join(ARTIFACTS_DIR, "x")}#ABCD:1-2]`;
+		expect(resolvePlanPath(session, nonHexHeader)).toBe(path.join(REPO_ROOT, nonHexHeader));
+		expect(resolvePlanPath(session, selectorHeader)).toBe(path.join(REPO_ROOT, selectorHeader));
 	});
 });
 
@@ -88,19 +87,19 @@ describe("enforcePlanModeWrite (working tree read-only, local:// sandbox writabl
 	const planMode: PlanModeState = { enabled: true, planFilePath: "local://some-plan.md" };
 
 	it("accepts writes to any local:// file", () => {
-		const session = makeSession({ artifactsDir: "/tmp/agent-artifacts", planMode });
+		const session = makeSession({ artifactsDir: ARTIFACTS_DIR, planMode });
 		expect(() => enforcePlanModeWrite(session, "local://auth-refactor-plan.md", { op: "create" })).not.toThrow();
 		expect(() => enforcePlanModeWrite(session, "local://scratch/notes.md", { op: "update" })).not.toThrow();
 	});
 
 	it("rejects writes to the working tree", () => {
-		const session = makeSession({ artifactsDir: "/tmp/agent-artifacts", cwd: "/repo", planMode });
+		const session = makeSession({ artifactsDir: ARTIFACTS_DIR, cwd: REPO_ROOT, planMode });
 		expect(() => enforcePlanModeWrite(session, "src/foo.ts", { op: "update" })).toThrow(/working tree is read-only/);
 		expect(() => enforcePlanModeWrite(session, "PLAN.md", { op: "create" })).toThrow(/working tree is read-only/);
 	});
 
 	it("rejects deletes and renames outright", () => {
-		const session = makeSession({ artifactsDir: "/tmp/agent-artifacts", planMode });
+		const session = makeSession({ artifactsDir: ARTIFACTS_DIR, planMode });
 		expect(() => enforcePlanModeWrite(session, "local://some-plan.md", { op: "delete" })).toThrow(
 			/deleting files is not allowed/,
 		);
@@ -110,7 +109,7 @@ describe("enforcePlanModeWrite (working tree read-only, local:// sandbox writabl
 	});
 
 	it("is a no-op when plan mode is disabled", () => {
-		const session = makeSession({ artifactsDir: "/tmp/agent-artifacts", cwd: "/repo" });
+		const session = makeSession({ artifactsDir: ARTIFACTS_DIR, cwd: REPO_ROOT });
 		expect(() => enforcePlanModeWrite(session, "src/foo.ts", { op: "update" })).not.toThrow();
 	});
 });
@@ -119,47 +118,57 @@ describe("enforcePlanModeWrite accepts absolute local-sandbox paths", () => {
 	const planMode: PlanModeState = { enabled: true, planFilePath: "local://some-plan.md" };
 
 	it("allows the absolute path returned by `read local://...` (== sandbox-resolved path)", async () => {
-		// Use an existing tmp directory so the realpath check inside the guard
-		// sees a real filesystem (macOS collapses /tmp -> /private/tmp etc.).
+		// Use an existing temp directory so the realpath check inside the guard
+		// sees a real filesystem even when the OS exposes temp paths through aliases.
 		const artifactsDir = await fs.mkdtemp(path.join(os.tmpdir(), "plan-guard-test-"));
-		const session = makeSession({ artifactsDir, planMode });
-		const absolute = resolvePlanPath(session, "local://my-plan.md");
-		expect(() => enforcePlanModeWrite(session, absolute, { op: "update" })).not.toThrow();
+		try {
+			const session = makeSession({ artifactsDir, planMode });
+			const absolute = resolvePlanPath(session, "local://my-plan.md");
+			expect(() => enforcePlanModeWrite(session, absolute, { op: "update" })).not.toThrow();
+		} finally {
+			await fs.rm(artifactsDir, { recursive: true, force: true });
+		}
 	});
 
 	it("allows bracketed hashline headers for local sandbox paths", async () => {
 		const artifactsDir = await fs.mkdtemp(path.join(os.tmpdir(), "plan-guard-test-"));
-		const session = makeSession({ artifactsDir, planMode });
-		const absolute = resolvePlanPath(session, "local://my-plan.md");
+		try {
+			const session = makeSession({ artifactsDir, planMode });
+			const absolute = resolvePlanPath(session, "local://my-plan.md");
 
-		// Strict hashline shape `[PATH]` or `[PATH#XXXX]` is unwrapped to the
-		// inner path for both the sandbox check and the eventual resolution.
-		expect(() => enforcePlanModeWrite(session, `[${absolute}#ABCD]`, { op: "update" })).not.toThrow();
-		expect(() => enforcePlanModeWrite(session, `[${absolute}]`, { op: "update" })).not.toThrow();
-		expect(() => enforcePlanModeWrite(session, `[local://my-plan.md#ABCD]`, { op: "update" })).not.toThrow();
+			// Strict hashline shape `[PATH]` or `[PATH#XXXX]` is unwrapped to the
+			// inner path for both the sandbox check and the eventual resolution.
+			expect(() => enforcePlanModeWrite(session, `[${absolute}#ABCD]`, { op: "update" })).not.toThrow();
+			expect(() => enforcePlanModeWrite(session, `[${absolute}]`, { op: "update" })).not.toThrow();
+			expect(() => enforcePlanModeWrite(session, `[local://my-plan.md#ABCD]`, { op: "update" })).not.toThrow();
+		} finally {
+			await fs.rm(artifactsDir, { recursive: true, force: true });
+		}
 	});
 
 	it("rejects malformed bracketed headers instead of silently unwrapping them", () => {
-		const session = makeSession({ artifactsDir: "/tmp/agent-artifacts", cwd: "/repo", planMode });
+		const session = makeSession({ artifactsDir: ARTIFACTS_DIR, cwd: REPO_ROOT, planMode });
+		const sandboxPlanPath = path.join(ARTIFACTS_DIR, "local", "plan.md");
 
 		// Selector tails (`#TAG:lines`), non-hex tags, and short tags fall outside
 		// the strict header shape; we leave them alone so the downstream resolver
 		// surfaces the real error rather than treating the bracketed blob as a path.
-		expect(() =>
-			enforcePlanModeWrite(session, "[/tmp/agent-artifacts/local/plan.md#ABCD:1-2]", { op: "update" }),
-		).toThrow(/working tree is read-only/);
-		expect(() =>
-			enforcePlanModeWrite(session, "[/tmp/agent-artifacts/local/plan.md#nothex]", { op: "update" }),
-		).toThrow(/working tree is read-only/);
+		expect(() => enforcePlanModeWrite(session, `[${sandboxPlanPath}#ABCD:1-2]`, { op: "update" })).toThrow(
+			/working tree is read-only/,
+		);
+		expect(() => enforcePlanModeWrite(session, `[${sandboxPlanPath}#nothex]`, { op: "update" })).toThrow(
+			/working tree is read-only/,
+		);
 	});
 
 	it("still rejects absolute paths outside the local sandbox", () => {
-		const session = makeSession({ artifactsDir: "/tmp/agent-artifacts", cwd: "/repo", planMode });
+		const session = makeSession({ artifactsDir: ARTIFACTS_DIR, cwd: REPO_ROOT, planMode });
+		const workingTreePath = path.join(REPO_ROOT, "src", "foo.ts");
 
-		expect(() => enforcePlanModeWrite(session, "/repo/src/foo.ts", { op: "update" })).toThrow(
+		expect(() => enforcePlanModeWrite(session, workingTreePath, { op: "update" })).toThrow(
 			/working tree is read-only/,
 		);
-		expect(() => enforcePlanModeWrite(session, "[/repo/src/foo.ts#ABCD]", { op: "update" })).toThrow(
+		expect(() => enforcePlanModeWrite(session, `[${workingTreePath}#ABCD]`, { op: "update" })).toThrow(
 			/working tree is read-only/,
 		);
 	});

packages/coding-agent/test/tools/read-renderer.test.ts

@@ -1,4 +1,6 @@
 import { afterAll, afterEach, beforeAll, describe, expect, it } from "bun:test";
+import * as path from "node:path";
+import * as url from "node:url";
 import { resetSettingsForTest, Settings, settings } from "@oh-my-pi/pi-coding-agent/config/settings";
 import { ToolExecutionComponent } from "@oh-my-pi/pi-coding-agent/modes/components/tool-execution";
 import { theme as activeTheme, getThemeByName, initTheme } from "@oh-my-pi/pi-coding-agent/modes/theme/theme";
@@ -35,11 +37,12 @@ describe("readToolRenderer hyperlinks", () => {
 		const theme = await getThemeByName("dark");
 		expect(theme).toBeDefined();
 
+		const handoffPath = path.resolve("/tmp/omp-local/handoff.md");
 		const component = readToolRenderer.renderResult(
 			{
 				content: [{ type: "text", text: "second line" }],
 				details: {
-					resolvedPath: "/tmp/omp-local/handoff.md",
+					resolvedPath: handoffPath,
 					displayContent: { text: "second line", startLine: 2 },
 					contentType: "text/plain",
 				},
@@ -52,7 +55,9 @@ describe("readToolRenderer hyperlinks", () => {
 		const rendered = component.render(200).join("\n");
 		expect(rendered).toContain("local://handoff.md");
 		expect(rendered).toContain(":2");
-		expect(extractLinkUris(rendered)).toContain("file:///tmp/omp-local/handoff.md?line=2");
+		const handoffUri = new URL(url.pathToFileURL(path.resolve(handoffPath)).href);
+		handoffUri.searchParams.set("line", "2");
+		expect(extractLinkUris(rendered)).toContain(handoffUri.href);
 		expect(extractLinkTexts(rendered)).toContain("local://handoff.md");
 		expect(extractLinkTexts(rendered)).not.toContain("local://handoff.md:2");
 	});
@@ -62,17 +67,20 @@ describe("readToolRenderer hyperlinks", () => {
 		const theme = await getThemeByName("dark");
 		expect(theme).toBeDefined();
 
+		const examplePath = path.resolve("/tmp/omp-read/example.ts");
 		const component = readToolRenderer.renderCall(
-			{ path: "/tmp/omp-read/example.ts:10-12" },
+			{ path: `${examplePath}:10-12` },
 			{ expanded: false, isPartial: false },
 			theme!,
 		);
 
 		const rendered = component.render(200).join("\n");
-		expect(Bun.stripANSI(rendered)).toContain("/tmp/omp-read/example.ts:10-12");
-		expect(extractLinkUris(rendered)).toContain("file:///tmp/omp-read/example.ts?line=10");
-		expect(extractLinkTexts(rendered)).toContain("/tmp/omp-read/example.ts");
-		expect(extractLinkTexts(rendered)).not.toContain("/tmp/omp-read/example.ts:10-12");
+		expect(Bun.stripANSI(rendered)).toContain(`${examplePath}:10-12`);
+		const exampleUri = new URL(url.pathToFileURL(path.resolve(examplePath)).href);
+		exampleUri.searchParams.set("line", "10");
+		expect(extractLinkUris(rendered)).toContain(exampleUri.href);
+		expect(extractLinkTexts(rendered)).toContain(examplePath);
+		expect(extractLinkTexts(rendered)).not.toContain(`${examplePath}:10-12`);
 	});
 
 	it("links HTTP read result headers to the final URL", async () => {

packages/coding-agent/test/tools/render-utils.test.ts

@@ -12,6 +12,7 @@ import {
 	formatExpandHint,
 	formatParseErrors,
 	formatScreenshot,
+	shortenPath,
 	truncateDiffByHunk,
 } from "@oh-my-pi/pi-coding-agent/tools/render-utils";
 import { getKeybindings, setKeybindings, type KeybindingsManager as TuiKeybindingsManager } from "@oh-my-pi/pi-tui";
@@ -101,7 +102,19 @@ describe("formatScreenshot", () => {
 		]);
 	});
 
+	it("uses forward slashes after a shortened Windows home", () => {
+		const home = String.raw`C:\Users\me`;
+		expect(shortenPath(String.raw`C:\Users\me\projects\demo`, home)).toBe("~/projects/demo");
+	});
+
+	it("does not shorten paths outside the home boundary", () => {
+		const home = String.raw`C:\Users\me`;
+		const sibling = String.raw`C:\Users\me2\projects\demo`;
+		expect(shortenPath(sibling, home)).toBe(sibling);
+	});
+
 	it("formats non-home path without tilde", () => {
+		const filePath = path.join(path.parse(os.homedir()).root, "omp-render-utils", "capture.png");
 		const resized = fakeResized({ mimeType: "image/webp", buffer: new Uint8Array(1024) });
 
 		expect(
@@ -109,12 +122,12 @@ describe("formatScreenshot", () => {
 				saveFullRes: true,
 				savedMimeType: "image/png",
 				savedByteLength: 2048,
-				dest: "/tmp/capture.png",
+				dest: filePath,
 				resized,
 			}),
 		).toEqual([
 			"Screenshot captured",
-			"Saved: image/png (2.00 KB) to /tmp/capture.png",
+			`Saved: image/png (2.00 KB) to ${filePath}`,
 			"Model: image/webp (1.00 KB, 800x600)",
 		]);
 	});
@@ -127,7 +140,7 @@ describe("formatScreenshot", () => {
 				saveFullRes: false,
 				savedMimeType: "image/webp",
 				savedByteLength: 3072,
-				dest: "/tmp/omp-sshots-123.png",
+				dest: path.join(os.tmpdir(), "omp-sshots-123.png"),
 				resized,
 			}),
 		).toEqual(["Screenshot captured", "Format: image/webp (3.00 KB)", "Dimensions: 800x600"]);
@@ -146,7 +159,7 @@ describe("formatScreenshot", () => {
 			saveFullRes: false,
 			savedMimeType: "image/webp",
 			savedByteLength: 2048,
-			dest: "/tmp/shot.png",
+			dest: path.join(os.tmpdir(), "shot.png"),
 			resized,
 		});
 

packages/coding-agent/test/tools/search-renderer.test.ts

@@ -1,4 +1,6 @@
 import { afterAll, afterEach, beforeAll, describe, expect, it } from "bun:test";
+import * as path from "node:path";
+import * as url from "node:url";
 import { resetSettingsForTest, Settings, settings } from "@oh-my-pi/pi-coding-agent/config/settings";
 import { getThemeByName } from "@oh-my-pi/pi-coding-agent/modes/theme/theme";
 import { searchToolRenderer } from "@oh-my-pi/pi-coding-agent/tools/search";
@@ -132,12 +134,14 @@ describe("searchToolRenderer", () => {
 		expect(theme).toBeDefined();
 		const uiTheme = theme!;
 
+		const projectRoot = path.resolve("/tmp/omp-project");
+		const filePath = path.join(projectRoot, "src", "file.ts");
 		const result = {
 			content: [{ type: "text", text: "" }],
 			details: {
 				matchCount: 1,
 				fileCount: 1,
-				searchPath: "/tmp/omp-project",
+				searchPath: projectRoot,
 				scopePath: "src",
 				displayContent: ["# src/", "## file.ts#abcd", "*12│const needle = true;"].join("\n"),
 			},
@@ -147,10 +151,13 @@ describe("searchToolRenderer", () => {
 			.renderResult(result as never, { expanded: true, isPartial: false }, uiTheme, { pattern: "needle" })
 			.render(240)
 			.join("\n");
+		const fileUri = url.pathToFileURL(path.resolve(filePath)).href;
+		const lineUri = new URL(fileUri);
+		lineUri.searchParams.set("line", "12");
 		const uris = extractLinkUris(rendered);
 
-		expect(uris).toContain("file:///tmp/omp-project/src/file.ts");
-		expect(uris).toContain("file:///tmp/omp-project/src/file.ts?line=12");
+		expect(uris).toContain(fileUri);
+		expect(uris).toContain(lineUri.href);
 	});
 
 	it("links single-file code-frame lines to the searched file", async () => {
@@ -159,12 +166,13 @@ describe("searchToolRenderer", () => {
 		expect(theme).toBeDefined();
 		const uiTheme = theme!;
 
+		const filePath = path.resolve("/tmp/omp-project/file.ts");
 		const result = {
 			content: [{ type: "text", text: "" }],
 			details: {
 				matchCount: 1,
 				fileCount: 1,
-				searchPath: "/tmp/omp-project/file.ts",
+				searchPath: filePath,
 				scopePath: "file.ts",
 				displayContent: "*7│needle();",
 			},
@@ -175,7 +183,9 @@ describe("searchToolRenderer", () => {
 			.render(240)
 			.join("\n");
 
-		expect(extractLinkUris(rendered)).toContain("file:///tmp/omp-project/file.ts?line=7");
+		const lineUri = new URL(url.pathToFileURL(path.resolve(filePath)).href);
+		lineUri.searchParams.set("line", "7");
+		expect(extractLinkUris(rendered)).toContain(lineUri.href);
 	});
 
 	it("bounds the expanded single-file view instead of dumping every match", async () => {
@@ -194,12 +204,13 @@ describe("searchToolRenderer", () => {
 			})
 			.join("\n");
 
+		const filePath = path.resolve("/tmp/omp-project/renderer.ts");
 		const result = {
 			content: [{ type: "text", text: "" }],
 			details: {
 				matchCount: clusters.length,
 				fileCount: 1,
-				searchPath: "/tmp/omp-project/renderer.ts",
+				searchPath: filePath,
 				scopePath: "renderer.ts",
 				displayContent,
 			},

packages/coding-agent/test/tools/tool-output-hyperlinks.test.ts

@@ -2,6 +2,7 @@ import { afterAll, afterEach, beforeAll, describe, expect, it } from "bun:test";
 import * as fs from "node:fs";
 import * as os from "node:os";
 import * as path from "node:path";
+import * as url from "node:url";
 import { resetSettingsForTest, Settings, settings } from "@oh-my-pi/pi-coding-agent/config/settings";
 import { editToolRenderer } from "@oh-my-pi/pi-coding-agent/edit/renderer";
 import { getThemeByName, initTheme } from "@oh-my-pi/pi-coding-agent/modes/theme/theme";
@@ -77,8 +78,8 @@ describe("tool output OSC 8 file:// hyperlinks", () => {
 				.render(200)
 				.join("\n");
 
-			expect(extractLinkUris(textRender)).toContain(`file://${textPath}`);
-			expect(extractLinkUris(imgRender)).toContain(`file://${imgPath}`);
+			expect(extractLinkUris(textRender)).toContain(url.pathToFileURL(path.resolve(textPath)).href);
+			expect(extractLinkUris(imgRender)).toContain(url.pathToFileURL(path.resolve(imgPath)).href);
 		} finally {
 			fs.rmSync(dir, { recursive: true, force: true });
 		}
@@ -101,7 +102,7 @@ describe("tool output OSC 8 file:// hyperlinks", () => {
 				)
 				.render(200)
 				.join("\n");
-			expect(extractLinkUris(rendered)).toContain(`file://${filePath}`);
+			expect(extractLinkUris(rendered)).toContain(url.pathToFileURL(path.resolve(filePath)).href);
 		} finally {
 			fs.rmSync(dir, { recursive: true, force: true });
 		}
@@ -113,13 +114,16 @@ describe("tool output OSC 8 file:// hyperlinks", () => {
 		// Scoped search: scope dir (`searchPath`) is below cwd, and the grouped
 		// display paths are cwd-relative. Resolving against searchPath would double
 		// the `src` prefix (`/proj/src/src/...`).
+		const projectRoot = path.resolve("/tmp/omp-project");
+		const srcRoot = path.join(projectRoot, "src");
+		const interactiveModePath = path.join(srcRoot, "interactive-mode.ts");
 		const result = {
 			content: [{ type: "text", text: "" }],
 			details: {
 				matchCount: 1,
 				fileCount: 1,
-				cwd: "/tmp/omp-project",
-				searchPath: "/tmp/omp-project/src",
+				cwd: projectRoot,
+				searchPath: srcRoot,
 				scopePath: "src",
 				displayContent: ["# src/", "## interactive-mode.ts#abcd", "*12│const needle = true;"].join("\n"),
 			},
@@ -128,15 +132,21 @@ describe("tool output OSC 8 file:// hyperlinks", () => {
 			.renderResult(result as never, { expanded: true, isPartial: false }, theme, { pattern: "needle" })
 			.render(240)
 			.join("\n");
+		const interactiveModeUri = url.pathToFileURL(path.resolve(interactiveModePath)).href;
+		const interactiveModeLineUri = new URL(interactiveModeUri);
+		interactiveModeLineUri.searchParams.set("line", "12");
 		const uris = extractLinkUris(rendered);
-		expect(uris).toContain("file:///tmp/omp-project/src/interactive-mode.ts");
-		expect(uris).toContain("file:///tmp/omp-project/src/interactive-mode.ts?line=12");
+		expect(uris).toContain(interactiveModeUri);
+		expect(uris).toContain(interactiveModeLineUri.href);
 		expect(uris.some(uri => uri.includes("/src/src/"))).toBe(false);
 	});
 
 	it("resolves scoped ast-grep links against cwd, not the (sub)scope path", async () => {
 		settings.override("tui.hyperlinks", "always");
 		const theme = (await getThemeByName("dark"))!;
+		const projectRoot = path.resolve("/tmp/omp-project");
+		const srcRoot = path.join(projectRoot, "src");
+		const interactiveModePath = path.join(srcRoot, "interactive-mode.ts");
 		const result = {
 			content: [{ type: "text", text: "" }],
 			details: {
@@ -144,8 +154,8 @@ describe("tool output OSC 8 file:// hyperlinks", () => {
 				fileCount: 1,
 				filesSearched: 1,
 				limitReached: false,
-				cwd: "/tmp/omp-project",
-				searchPath: "/tmp/omp-project/src",
+				cwd: projectRoot,
+				searchPath: srcRoot,
 				scopePath: "src",
 				displayContent: ["# src/", "## interactive-mode.ts", "  *12│const needle = true;"].join("\n"),
 			},
@@ -154,19 +164,21 @@ describe("tool output OSC 8 file:// hyperlinks", () => {
 			.renderResult(result as never, { expanded: true, isPartial: false }, theme, { pat: "needle" })
 			.render(240)
 			.join("\n");
+		const interactiveModeUri = url.pathToFileURL(path.resolve(interactiveModePath)).href;
 		const uris = extractLinkUris(rendered);
-		expect(uris).toContain("file:///tmp/omp-project/src/interactive-mode.ts");
+		expect(uris).toContain(interactiveModeUri);
 		expect(uris.some(uri => uri.includes("/src/src/"))).toBe(false);
 	});
 
 	it("links the edit header to the absolute details.path even when the arg path is relative", async () => {
 		settings.override("tui.hyperlinks", "always");
 		const theme = (await getThemeByName("dark"))!;
+		const editPath = path.resolve("/tmp/omp-project/src/a.ts");
 		const rendered = editToolRenderer
 			.renderResult(
 				{
 					content: [{ type: "text", text: "Updated src/a.ts" }],
-					details: { diff: "+1|// x", op: "update", path: "/tmp/omp-project/src/a.ts" },
+					details: { diff: "+1|// x", op: "update", path: editPath },
 				},
 				{ expanded: false, isPartial: false, renderContext: { editMode: "hashline" } },
 				theme,
@@ -174,9 +186,11 @@ describe("tool output OSC 8 file:// hyperlinks", () => {
 			)
 			.render(200)
 			.join("\n");
+		const editUri = url.pathToFileURL(path.resolve(editPath)).href;
+		const rootAnchoredArgUri = url.pathToFileURL(path.resolve("/src/a.ts")).href;
 		const uris = extractLinkUris(rendered);
-		expect(uris).toContain("file:///tmp/omp-project/src/a.ts");
-		// A relative arg path must not leak into a root-anchored `file:///src/a.ts`.
-		expect(uris).not.toContain("file:///src/a.ts");
+		expect(uris).toContain(editUri);
+		// A relative arg path must not leak into a root-anchored file URI.
+		expect(uris).not.toContain(rootAnchoredArgUri);
 	});
 });

packages/coding-agent/test/tui/hyperlink.test.ts

@@ -1,4 +1,6 @@
 import { afterAll, afterEach, beforeAll, beforeEach, describe, expect, it } from "bun:test";
+import * as path from "node:path";
+import * as url from "node:url";
 import { resetSettingsForTest, Settings, settings } from "@oh-my-pi/pi-coding-agent/config/settings";
 import { LocalProtocolHandler } from "@oh-my-pi/pi-coding-agent/internal-urls/local-protocol";
 import { AgentRegistry } from "@oh-my-pi/pi-coding-agent/registry/agent-registry";
@@ -112,20 +114,23 @@ describe("isHyperlinkEnabled", () => {
 describe("fileHyperlink", () => {
 	it("returns plain text when hyperlinks are disabled (mode=off)", () => {
 		setHyperlinkMode("off");
-		const result = fileHyperlink("/Users/foo/bar.ts", "bar.ts");
+		const filePath = path.resolve("/Users/foo/bar.ts");
+		const result = fileHyperlink(filePath, "bar.ts");
 		expect(result).toBe("bar.ts");
 	});
 
 	it("wraps text in OSC 8 when hyperlinks are enabled (mode=always)", () => {
 		setHyperlinkMode("always");
-		const result = fileHyperlink("/Users/foo/bar.ts", "bar.ts");
+		const filePath = path.resolve("/Users/foo/bar.ts");
+		const result = fileHyperlink(filePath, "bar.ts");
 		expect(isHyperlinked(result)).toBe(true);
 		expect(result).toContain("bar.ts");
 	});
 
 	it("builds a valid file:// URI with the absolute path", () => {
 		setHyperlinkMode("always");
-		const result = fileHyperlink("/Users/foo/bar.ts", "bar.ts");
+		const filePath = path.resolve("/Users/foo/bar.ts");
+		const result = fileHyperlink(filePath, "bar.ts");
 		const uri = extractLinkUri(result);
 		expect(uri).toMatch(/^file:\/\//);
 		expect(uri).toContain("bar.ts");
@@ -133,7 +138,8 @@ describe("fileHyperlink", () => {
 
 	it("encodes spaces in the path", () => {
 		setHyperlinkMode("always");
-		const result = fileHyperlink("/Users/foo/my file.ts", "my file.ts");
+		const filePath = path.resolve("/Users/foo/my file.ts");
+		const result = fileHyperlink(filePath, "my file.ts");
 		const uri = extractLinkUri(result);
 		expect(uri).toContain("%20");
 		expect(uri).not.toContain(" ");
@@ -141,9 +147,12 @@ describe("fileHyperlink", () => {
 
 	it("percent-encodes URL-reserved path bytes before appending query params", () => {
 		setHyperlinkMode("always");
-		const result = fileHyperlink("/Users/foo/a#b?c% d.ts", "a#b?c% d.ts", { line: 12 });
+		const filePath = path.resolve("/Users/foo/a#b?c% d.ts");
+		const result = fileHyperlink(filePath, "a#b?c% d.ts", { line: 12 });
 		const uri = extractLinkUri(result);
-		expect(uri).toBe("file:///Users/foo/a%23b%3Fc%25%20d.ts?line=12");
+		const expectedUri = new URL(url.pathToFileURL(path.resolve(filePath)).href);
+		expectedUri.searchParams.set("line", "12");
+		expect(uri).toBe(expectedUri.href);
 	});
 
 	it("resolves relative paths before building file URIs", () => {
@@ -156,7 +165,8 @@ describe("fileHyperlink", () => {
 
 	it("appends line and col as query params when provided", () => {
 		setHyperlinkMode("always");
-		const result = fileHyperlink("/Users/foo/bar.ts", "bar.ts", { line: 42, col: 7 });
+		const filePath = path.resolve("/Users/foo/bar.ts");
+		const result = fileHyperlink(filePath, "bar.ts", { line: 42, col: 7 });
 		const uri = extractLinkUri(result);
 		expect(uri).toContain("line=42");
 		expect(uri).toContain("col=7");
@@ -164,15 +174,17 @@ describe("fileHyperlink", () => {
 
 	it("omits query params when line/col are not provided", () => {
 		setHyperlinkMode("always");
-		const result = fileHyperlink("/Users/foo/bar.ts", "bar.ts");
+		const filePath = path.resolve("/Users/foo/bar.ts");
+		const result = fileHyperlink(filePath, "bar.ts");
 		const uri = extractLinkUri(result);
 		expect(uri).not.toContain("?");
 	});
 
 	it("produces a stable id for the same path", () => {
 		setHyperlinkMode("always");
-		const r1 = fileHyperlink("/Users/foo/bar.ts", "bar.ts");
-		const r2 = fileHyperlink("/Users/foo/bar.ts", "different display text");
+		const filePath = path.resolve("/Users/foo/bar.ts");
+		const r1 = fileHyperlink(filePath, "bar.ts");
+		const r2 = fileHyperlink(filePath, "different display text");
 		// Extract id= from params (between "id=" and next ";")
 		const id1 = r1.match(/id=([^;]+)/)?.[1];
 		const id2 = r2.match(/id=([^;]+)/)?.[1];
@@ -182,8 +194,9 @@ describe("fileHyperlink", () => {
 
 	it("does not double-wrap text that already contains an OSC 8 sequence", () => {
 		setHyperlinkMode("always");
-		const alreadyWrapped = `${OSC}8;id=abc123;file:///foo/bar.ts${ST}bar.ts${LINK_END}`;
-		const result = fileHyperlink("/Users/foo/other.ts", alreadyWrapped);
+		const alreadyWrappedUri = url.pathToFileURL(path.resolve("/foo/bar.ts")).href;
+		const alreadyWrapped = `${OSC}8;id=abc123;${alreadyWrappedUri}${ST}bar.ts${LINK_END}`;
+		const result = fileHyperlink(path.resolve("/Users/foo/other.ts"), alreadyWrapped);
 		// Should return the already-wrapped text unchanged
 		expect(result).toBe(alreadyWrapped);
 	});
@@ -191,7 +204,8 @@ describe("fileHyperlink", () => {
 	it("preserves ANSI color codes inside the hyperlink", () => {
 		setHyperlinkMode("always");
 		const colored = "\x1b[32mbar.ts\x1b[0m";
-		const result = fileHyperlink("/Users/foo/bar.ts", colored);
+		const filePath = path.resolve("/Users/foo/bar.ts");
+		const result = fileHyperlink(filePath, colored);
 		expect(result).toContain(colored);
 		expect(isHyperlinked(result)).toBe(true);
 	});

packages/coding-agent/test/utils/filter-user-extensions.ts

@@ -31,7 +31,11 @@ function lexicalIsWithin(root: string, candidate: string): boolean {
 	return relative === "" || (!relative.startsWith("..") && !path.isAbsolute(relative));
 }
 
-export function filterUserScoped<T extends { path: string }>(items: T[]): T[] {
+export function filterUserScoped<T extends { path: string }>(items: T[], keepRoots?: string | string[]): T[] {
+	if (keepRoots) {
+		const roots = Array.isArray(keepRoots) ? keepRoots : [keepRoots];
+		return items.filter(it => roots.some(root => lexicalIsWithin(root, it.path)));
+	}
 	const prefixes = [getConfigRootDir(), getAgentDir(), getPluginsDir()];
 	return items.filter(it => !prefixes.some(prefix => lexicalIsWithin(prefix, it.path)));
 }

packages/coding-agent/test/utils/git-clone.test.ts

@@ -2,6 +2,7 @@ import { afterAll, beforeAll, describe, expect, test } from "bun:test";
 import * as fs from "node:fs/promises";
 import * as os from "node:os";
 import * as path from "node:path";
+import * as url from "node:url";
 
 import * as git from "@oh-my-pi/pi-coding-agent/utils/git";
 
@@ -44,7 +45,7 @@ describe("git.clone with options.sha", () => {
 
 		// `file://` is required: local-path clones ignore `--depth`, which would
 		// mask the bug. See git-clone(1) "GIT URLS" / "LOCAL PROTOCOL".
-		upstreamUrl = `file://${upstream}`;
+		upstreamUrl = url.pathToFileURL(upstream).href;
 
 		gitRun(upstream, ["init", "-q", "-b", "main"]);
 		gitRun(upstream, ["commit", "-q", "--allow-empty", "-m", "first"]);

packages/utils/CHANGELOG.md

@@ -2,12 +2,15 @@
 
 ## [Unreleased]
 
+### Changed
+
+- Expanded the `TempDir` Windows retry window from 4×10ms to 40×25ms (1s total) to accommodate SQLite WAL/SHM file handle release delays
+
 ## [16.0.11] - 2026-06-19
 
 ### Removed
 
 - Removed `getIndentation`, `setDefaultTabWidth`, and `getDefaultTabWidth` helpers
-
 ## [16.0.8] - 2026-06-18
 
 ### Changed

packages/utils/src/temp.ts

@@ -78,8 +78,8 @@ function normalizePrefix(prefix?: string): string {
 }
 
 const kRemoveOptions = { recursive: true, force: true } as const;
-const kRemoveRetries = 4;
-const kRemoveRetryDelayMs = 10;
+const kRemoveRetries = 40;
+const kRemoveRetryDelayMs = 25;
 const kRetryableRemoveErrorCodes = new Set(["EBUSY", "EPERM", "ENOTEMPTY"]);
 const kSleepBuffer = new Int32Array(new SharedArrayBuffer(4));