feat(coding-agent): removed context argument from eval agent() spawn

Shared background now flows through a '/Users/can/.omp/agent/sessions/-Projects-.tree-pi-commit/2026-06-10T15-36-32-782Z_019eb22d-970e-7000-8964-72c98becf3e8/local' file referenced in each prompt instead of a context string forwarded into the subagent's system prompt. The JS and Python preludes drop the context kwarg from agent(), the subagent system prompt drops the {{#if context}} block and the conversation-context file pointer, and runEvalAgent no longer writes a per-call conversation context file. AgentSession sheds the now-unused formatCompactContext() helper that supplied the file's body, and ToolSession.getCompactContext is removed alongside it.

docs/tools/eval.md

@@ -138,7 +138,7 @@ Implemented in `packages/coding-agent/src/eval/js/worker-core.ts`, `packages/cod
   - `await read(path, { offset?, limit? })`
   - `await tree(path = ".", { maxDepth?, hidden? })`
   - `sort(text, { reverse?, unique? })`, `uniq(text, { count? })`, `counter(items, { limit?, reverse? })`
-  - `await agent(prompt, { agentType?, model?, context?, label?, schema? })`
+  - `await agent(prompt, { agentType?, model?, label?, schema? })`
   - `await parallel([() => agent("a"), () => agent("b")])`
   - `await pipeline(items, stage1, stage2)`
 - `display(value)` behavior:
@@ -192,11 +192,11 @@ Both runtimes expose `completion()` — a single stateless completion against a
 Both runtimes expose `agent()` — a single subagent invocation routed through `packages/coding-agent/src/eval/agent-bridge.ts` into the same `runSubprocess(...)` path used by the `task` tool. It uses the current eval session's spawn policy and inherits the parent eval executor id, so parent and subagent code share JS/Python runtime state.
 
 - Signatures:
-  - JS: `await agent(prompt, { agentType?, model?, context?, label?, schema? })`
+  - JS: `await agent(prompt, { agentType?, model?, label?, schema? })`
-  - Python: `agent(prompt, *, agent_type="task", model=None, context=None, label=None, schema=None)`
+  - Python: `agent(prompt, *, agent_type="task", model=None, label=None, schema=None)`
 - `agentType` / `agent_type` defaults to the bundled `task` agent and resolves through normal agent discovery, so project and user agents work.
 - `model` overrides the selected agent's model. Without it, normal per-agent settings and the agent frontmatter model apply.
-- `context` supplies shared background; `label` controls the `agent://<id>` output label prefix.
+- Shared background is passed via files: write a `local://` file and reference it in the prompt. `label` controls the `agent://<id>` output label prefix.
 - `schema` passes a JSON Schema to the subagent structured-output path. When present, the helper parses the final JSON text and returns an object.
 - Spawn restrictions use `session.getSessionSpawns()` exactly like the `task` tool. Eval-driven subagent recursion is capped at depth 3.
 - JS and Python both expose `parallel(thunks)` and `pipeline(items, ...stages)`; both use a bounded async/threaded pool whose width tracks the `task.maxConcurrency` setting (the same ceiling the `task` tool uses; `0` = run every item at once), preserve item order, and propagate rejections. The width is fetched live from the host via the `__concurrency__` bridge, so the helpers no longer take a `concurrency` argument.

packages/coding-agent/src/eval/__tests__/agent-bridge.test.ts

@@ -178,7 +178,7 @@ describe("runEvalAgent", () => {
 		expect(runSpy).not.toHaveBeenCalled();
 	});
 
-	it("passes the parent execution context and only sets outputSchema when schema is supplied", async () => {
+	it("passes parent execution options and only sets outputSchema when schema is supplied", async () => {
 		mockAgents();
 		const runSpy = vi.spyOn(taskExecutor, "runSubprocess").mockImplementation(async options => singleResult(options));
 		const abortController = new AbortController();
@@ -186,7 +186,7 @@ describe("runEvalAgent", () => {
 		const session = makeSession({ depth: 2, activeModel: "p/current", modelString: "p/fallback" });
 
 		await runEvalAgent(
-			{ prompt: " hello ", context: " context ", label: "My Agent", model: "p/override", schema },
+			{ prompt: " hello ", label: "My Agent", model: "p/override", schema },
 			{ session, signal: abortController.signal },
 		);
 		await runEvalAgent({ prompt: "plain" }, { session });
@@ -199,7 +199,6 @@ describe("runEvalAgent", () => {
 		expect(firstOptions.parentActiveModelPattern).toBe("p/current");
 		expect(firstOptions.outputSchema).toBe(schema);
 		expect(firstOptions.assignment).toBe("hello");
-		expect(firstOptions.context).toBe("context");
 		expect(firstOptions.description).toBe("My Agent");
 		expect(firstOptions.modelOverride).toEqual(["p/override"]);
 		expect(secondOptions.outputSchema).toBeUndefined();

packages/coding-agent/src/eval/agent-bridge.ts

@@ -34,7 +34,6 @@ const agentArgsSchema = z.object({
 	prompt: z.string().min(1, "prompt must be a non-empty string"),
 	agentType: z.string().min(1).optional(),
 	model: z.union([z.string().min(1), z.array(z.string().min(1)).min(1)]).optional(),
-	context: z.string().optional(),
 	label: z.string().optional(),
 	schema: z.unknown().optional(),
 });
@@ -43,7 +42,6 @@ interface EvalAgentArgs {
 	prompt: string;
 	agentType?: string;
 	model?: string | string[];
-	context?: string;
 	label?: string;
 	schema?: unknown;
 }
@@ -135,20 +133,12 @@ function getOutputManager(session: ToolSession): AgentOutputManager {
 async function getArtifacts(session: ToolSession): Promise<{
 	sessionFile: string | null;
 	artifactsDir: string;
-	contextFile?: string;
 }> {
 	const sessionFile = session.getSessionFile();
 	const sessionArtifactsDir = sessionFile ? sessionFile.slice(0, -6) : null;
 	const artifactsDir = sessionArtifactsDir ?? path.join(os.tmpdir(), `omp-eval-agent-${Snowflake.next()}`);
 	await fs.mkdir(artifactsDir, { recursive: true });
-
+	return { sessionFile, artifactsDir };
-	const shouldWriteConversationContext = session.settings.get("irc.enabled") !== true;
-	const compactContext = shouldWriteConversationContext ? session.getCompactContext?.() : undefined;
-	if (!compactContext) return { sessionFile, artifactsDir };
-
-	const contextFile = path.join(artifactsDir, "context.md");
-	await Bun.write(contextFile, compactContext);
-	return { sessionFile, artifactsDir, contextFile };
 }
 
 function emitProgressStatus(emitStatus: ((event: JsStatusEvent) => void) | undefined, progress: AgentProgress): void {
@@ -246,11 +236,10 @@ export async function runEvalAgent(args: unknown, options: EvalAgentBridgeOption
 	};
 	const parentArtifactManager = options.session.getArtifactManager?.() ?? undefined;
 	const mcpManager = options.session.mcpManager ?? MCPManager.instance();
-	const { sessionFile, artifactsDir, contextFile } = await getArtifacts(options.session);
+	const { sessionFile, artifactsDir } = await getArtifacts(options.session);
 	const outputManager = getOutputManager(options.session);
 	const id = await outputManager.allocate(outputIdBase(parsed.label, agentName));
 	const assignment = parsed.prompt.trim();
-	const context = trimToUndefined(parsed.context);
 	// Suspend eval timeout accounting while the subagent owns control. The
 	// timeout clock restarts once the bridge returns to the cell runtime.
 	const result = await withBridgeTimeoutPause(options.emitStatus, () =>
@@ -259,7 +248,6 @@ export async function runEvalAgent(args: unknown, options: EvalAgentBridgeOption
 			agent: effectiveAgent,
 			task: renderSubagentPrompt(assignment),
 			assignment,
-			context,
 			description: trimToUndefined(parsed.label),
 			index: 0,
 			id,
@@ -271,7 +259,6 @@ export async function runEvalAgent(args: unknown, options: EvalAgentBridgeOption
 			sessionFile,
 			persistArtifacts: Boolean(sessionFile),
 			artifactsDir,
-			contextFile,
 			// Eval `agent()` subagents are short-lived programmatic helpers (data
 			// collection, structured output, parallel() fan-out). LSP server
 			// cold-start costs tens of seconds and is pure overhead here, so it is

packages/coding-agent/src/eval/js/shared/prelude.txt

@@ -65,7 +65,7 @@ if (!globalThis.__omp_js_prelude_loaded__) {
 	};
 
 	const agent = async (prompt, opts, ...rest) => {
-		const o = optionsArg("agent", opts, rest, "{ agentType, model, context, label, schema }");
+		const o = optionsArg("agent", opts, rest, "{ agentType, model, label, schema }");
 		const res = await globalThis.__omp_call_tool__("__agent__", { prompt, ...o });
 		const text = res && typeof res === "object" ? res.text : res;
 		return hasOwn(o, "schema") ? JSON.parse(text) : text;

packages/coding-agent/src/eval/py/prelude.py

@@ -519,21 +519,20 @@ if "__omp_prelude_loaded__" not in globals():
         text = res.get("text") if isinstance(res, dict) else res
         return json.loads(text) if schema is not None else text
 
-    def agent(prompt, *, agent_type="task", model=None, context=None, label=None, schema=None):
+    def agent(prompt, *, agent_type="task", model=None, label=None, schema=None):
         """Run a subagent and return its final output.
 
         `agent_type` selects the subagent definition (default "task"). Pass
-        `model` to override that agent's model, `context` for shared background,
+        `model` to override that agent's model, `label` for the output artifact
-        `label` for the output artifact id, and `schema` to request structured
+        id, and `schema` to request structured JSON output; when `schema` is
-        JSON output; when `schema` is supplied the parsed object is returned.
+        supplied the parsed object is returned. Share background by writing a
+        local:// file and referencing it in the prompt.
         """
         args = {"prompt": prompt}
         if agent_type is not None:
             args["agentType"] = agent_type
         if model is not None:
             args["model"] = model
-        if context is not None:
-            args["context"] = context
         if label is not None:
             args["label"] = label
         if schema is not None:

packages/coding-agent/src/prompts/system/subagent-system-prompt.md

@@ -3,13 +3,6 @@ ROLE
 
 {{agent}}
 
-{{#if context}}
-CONTEXT
-===================================
-
-{{context}}
-{{/if}}
-
 {{#if planReference}}
 PLAN
 ===================================
@@ -32,11 +25,6 @@ You are working in an isolated working tree at `{{worktree}}` for this sub-task.
 You NEVER modify files outside this tree or in the original repository.
 {{/if}}
 
-{{#if contextFile}}
-# Conversation Context
-If you need additional information, your conversation with the user is in {{contextFile}} — `read` its tail or `search` it for relevant terms.
-{{/if}}
-
 {{#if ircPeers}}
 # IRC Peers
 You can reach other live agents via the `irc` tool. Your id is `{{ircSelfId}}`. Currently visible peers:

packages/coding-agent/src/prompts/system/workflow-notice.md

@@ -13,8 +13,8 @@ Worth it when the task benefits from decomposition + parallel coverage, or from
 <helpers>
 State persists across cells, so scout in one cell and fan out in the next. Every cell has:
 
-- `agent(prompt, *, agent_type="task", model=None, context=None, label=None, schema=None)` — run ONE subagent; returns its final text, or the validated object when `schema` (a JSON Schema dict) is given. With `schema` the subagent is forced to emit structured output that is validated for you — branch on the object, not on parsed prose. `agent_type` picks a discovered agent ("explore", "reviewer", "oracle", …); `context` is shared background; `label` names the artifact. Subagents are told their final text IS the return value, so they hand back raw data. `agent()` blocks until the subagent finishes; eval-spawned agents nest at most 3 deep.
+- `agent(prompt, *, agent_type="task", model=None, label=None, schema=None)` — run ONE subagent; returns its final text, or the validated object when `schema` (a JSON Schema dict) is given. With `schema` the subagent is forced to emit structured output that is validated for you — branch on the object, not on parsed prose. `agent_type` picks a discovered agent ("explore", "reviewer", "oracle", …); `label` names the artifact. Shared background goes in a `local://` file referenced from each prompt, not a parameter. Subagents are told their final text IS the return value, so they hand back raw data. `agent()` blocks until the subagent finishes; eval-spawned agents nest at most 3 deep.
-- `parallel(thunks)` — run zero-arg callables concurrently through a bounded pool, preserving input order; returns once all finish. The pool runs as wide as a `task` tool batch — don't hand-tune it; fan out as wide as the work divides. A thunk that raises propagates — wrap risky work in `try/except` inside the thunk to keep partial results. In a loop, bind each closure's value with a default arg (`lambda d=d: …`) or every thunk captures the last one.
+- `parallel(thunks)` — run zero-arg callables concurrently through a bounded pool, preserving input order; returns once all finish. The pool is bounded by the session's `task` concurrency — don't hand-tune it; fan out as wide as the work divides. A thunk that raises propagates — wrap risky work in `try/except` inside the thunk to keep partial results. In a loop, bind each closure's value with a default arg (`lambda d=d: …`) or every thunk captures the last one.
 - `pipeline(items, *stages)` — map items through `stages` left-to-right. There is a BARRIER between stages: ALL items clear stage N before stage N+1 begins. Each stage is a one-arg callable; stage 1 gets the original item, later stages get the previous result. Same pool width as `parallel()`.
 - `completion(prompt, *, model="default", system=None, schema=None)` — oneshot, stateless model call (no tools, no history). Tiers: "smol", "default", "slow". Cheap classification/scoring inside a fan-out.
 - `log(message)` — emit a progress line above the status tree. `phase(title)` — start a phase; the status lines that follow group under it.

packages/coding-agent/src/prompts/tools/eval.md

@@ -46,9 +46,9 @@ tool.<name>(args) → unknown
     Invoke any session tool by name. `args` is the tool's parameter object.
 completion(prompt, model?="default", system?=None, schema?=None) → str | dict
     Oneshot, stateless completion (no history, no tools). `model` picks a tier: "smol" (fast), "default" (this session's model), "slow" (most capable). Pass `system` for a system prompt. Pass a JSON-Schema `schema` to force structured output and get the parsed object back; otherwise returns the completion text.
-{{#if spawns}}agent(prompt, agent_type?="task", model?=None, context?=None, label?=None, schema?=None) → str | dict
+{{#if spawns}}agent(prompt, agent_type?="task", model?=None, label?=None, schema?=None) → str | dict
-    Run a subagent and return its final output. Defaults to the bundled "task" agent; pass `agent_type`/`agentType` for another discovered agent. Pass a JSON-Schema `schema` to force structured output and get the parsed object back.
+    Run a subagent and return its final output. Defaults to the bundled "task" agent; pass `agent_type`/`agentType` for another discovered agent. Pass a JSON-Schema `schema` to force structured output and get the parsed object back. Share background by writing a `local://` file and referencing it in the prompt.
-{{#if js}}    In JS, pass options as one trailing object — never positional: agent(prompt, { agentType, context, schema }).
+{{#if js}}    In JS, pass options as one trailing object — never positional: agent(prompt, { agentType, schema }).
 {{/if}}
 {{/if}}
 parallel(thunks) → list

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

@@ -231,10 +231,8 @@ import type { AuthStorage } from "./auth-storage";
 import type { ClientBridge, ClientBridgePermissionOption, ClientBridgePermissionOutcome } from "./client-bridge";
 import {
 	type BashExecutionMessage,
-	type CompactionSummaryMessage,
 	type CustomMessage,
 	convertToLlm,
-	type FileMentionMessage,
 	type PythonExecutionMessage,
 	readPendingDisplayTag,
 	SILENT_ABORT_MARKER,
@@ -9960,69 +9958,6 @@ export class AgentSession {
 		});
 	}
 
-	/**
-	 * Format the conversation as compact context for subagents.
-	 * Includes only user messages and assistant text responses.
-	 * Excludes: system prompt, tool definitions, tool calls/results, thinking blocks.
-	 */
-	formatCompactContext(): string {
-		const lines: string[] = [];
-		lines.push("# Conversation Context");
-		lines.push("");
-		lines.push(
-			"This is a summary of the parent conversation. Read this if you need additional context about what was discussed or decided.",
-		);
-		lines.push("");
-
-		for (const msg of this.messages) {
-			if (msg.role === "user" || msg.role === "developer") {
-				lines.push(msg.role === "developer" ? "## Developer" : "## User");
-				lines.push("");
-				if (typeof msg.content === "string") {
-					lines.push(msg.content);
-				} else {
-					for (const c of msg.content) {
-						if (c.type === "text") {
-							lines.push(c.text);
-						} else if (c.type === "image") {
-							lines.push("[Image attached]");
-						}
-					}
-				}
-				lines.push("");
-			} else if (msg.role === "assistant") {
-				const assistantMsg = msg as AssistantMessage;
-				// Only include text content, skip tool calls and thinking
-				const textParts: string[] = [];
-				for (const c of assistantMsg.content) {
-					if (c.type === "text" && c.text.trim()) {
-						textParts.push(c.text);
-					}
-				}
-				if (textParts.length > 0) {
-					lines.push("## Assistant");
-					lines.push("");
-					lines.push(textParts.join("\n\n"));
-					lines.push("");
-				}
-			} else if (msg.role === "fileMention") {
-				const fileMsg = msg as FileMentionMessage;
-				const paths = fileMsg.files.map(f => f.path).join(", ");
-				lines.push(`[Files referenced: ${paths}]`);
-				lines.push("");
-			} else if (msg.role === "compactionSummary") {
-				const compactMsg = msg as CompactionSummaryMessage;
-				lines.push("## Earlier Context (Summarized)");
-				lines.push("");
-				lines.push(compactMsg.summary);
-				lines.push("");
-			}
-			// Skip: toolResult, bashExecution, pythonExecution, branchSummary, custom, hookMessage
-		}
-
-		return lines.join("\n").trim();
-	}
-
 	// =========================================================================
 	// Extension System
 	// =========================================================================

packages/coding-agent/src/tools/index.ts

@@ -260,8 +260,6 @@ export interface ToolSession {
 	recordEvalSubagentUsage?: (output: number) => void;
 	/** Bridge to the connected client (e.g. ACP editor host). Tools should route fs/terminal/permission requests through this when available. */
 	getClientBridge?: () => ClientBridge | undefined;
-	/** Get compact conversation context for subagents (excludes tool results, system prompts) */
-	getCompactContext?: () => string;
 	/** Get cached todo phases for this session. */
 	getTodoPhases?: () => TodoPhase[];
 	/** Replace cached todo phases for this session. */