- Changed output format to group results under `# /` headers to reduce token usage for shared path prefixes.
- Clamped the `limit` parameter to 1-200 (default 200) instead of the previous 1000.
- Updated tests to assert against raw file lists instead of parsed text output.
- Parser now buffers markdown-style `#` lines and skips them when they directly precede a hashline operation.
- It now preserves comment lines that are not immediately before an operation while still handling blank separators as regular raw input.
- Added focused parser tests plus prompt and changelog updates describing the new comment-skipping behavior.
- Added a dedicated @oh-my-pi/hashline package with parser, patcher, filesystem, snapshots, and release metadata.
- Migrated coding-agent hashline and stream entrypoints to @oh-my-pi/hashline and removed old hashline module exports.
- Changed multi-section hashline execution to validate section hashes and flush diagnostics only at the final commit.
- Added session fileSnapshotStore support and rewired edit/read/search/write tools to use it instead of fileReadCache.
- Added `unfoldUntilLines`/`unfoldLimitLines` options to progressively reveal nested elidable spans breadth-first instead of collapsing everything behind the outermost elision.
- Added `minTotalLines` setting to skip summarization for short files, returning verbatim content instead.
- Added `:` selector support to `search` paths for constraining matches to specific line ranges.
- Extracted `parseLineRanges`/`parseLineRangeChunk`/`isLineInRanges` from `read.ts` into shared `path-utils.ts`.
- Measured bash wall-clock duration for direct, terminal-bridge, and interactive execution paths.
- Recorded wall time in result notices and details, then stripped the duplicated literal notice during shell rendering.
- Updated the renderer to include wall time in the status label and added tests for the new wall-time behavior.
- In `resolveApproval`, yolo mode now returns the user policy directly (`allow`/`prompt`/`deny`) and ignores tool `override` prompts.
- Updated approval-mode and approval unit tests to match the new behavior for critical bash patterns under yolo and auto-approve.
- Updated docs and settings metadata to describe yolo as user-policy-driven rather than override-driven.
- Added `ToolTier`, `ToolApproval`, and `ToolApprovalDecision` types and exported approval APIs.
- Updated approval-mode options from `auto|prompt|custom` to `always-ask|write|yolo` and defaulted mode to `yolo`.
- Changed approval resolution to apply per-tool decisions first, then mode-tier limits, with legacy-mode migration.
- Assigned read/write/exec `approval` and approval-detail prompts across built-in, custom, extension, and MCP tools.
- Decouple the per-tool approval gate from extension presence. ExtensionRunner
and the ExtensionToolWrapper that hosts the gate are now constructed
unconditionally in createAgentSession. Previously the runner was only built
when extensionsResult.extensions.length > 0, so the entire approval system
silently disappeared for sessions with no extensions loaded — any
tools.approvalMode: prompt|custom setting was a no-op without feedback.
Today this hole was masked by createAutoresearchExtension always being
pushed inline; the unconditional construction makes the safety invariant
explicit, and a new regression test in approval-mode.test.ts pins it.
- Extend CRITICAL_BASH_PATTERNS to cover remote-fetch-then-execute shapes
that the original `bash <(curl …)` regex missed:
- `source <(curl …)` / `. <(curl …)` (anchored at command boundary so
`find . -name foo` doesn't false-positive)
- `eval "$(curl …)"` / `eval $(curl …)` / `eval `curl …``
Also adds `chmod -R` symbolic-mode forms (`u+x`, `u+rwx,o+w …`) targeting
filesystem root, and `tee` / `tee -a` writes to /etc/{passwd,shadow,sudoers}
(the standard way to write root-owned files without redirect). Benign
forms (`source ./local.sh`, `chmod -R u+x ./build`, `tee /var/log/app.log`,
`eval "$VAR"`) are pinned negative in the test suite.
- Extend formatApprovalPrompt with payload previews for the destructive tools
that previously rendered as bare `Allow tool: <name>`: eval (language +
first cell's code), task (agent + first task's id + assignment), ast_edit
(first op's pattern / replacement / paths), browser (action + tab + url +
code), and write content (alongside path). For `task` in particular this
closes the gap that docs/approval-mode.md's "parent's approval covers the
subagent" claim was waving at — the prompt now actually shows what's being
delegated.
- Tighten isMcpToolName: drop the fallback `|| toolName.includes("__")` so
an extension tool legally named `my__feature` or `pkg__util__do` is no
longer falsely labelled `Origin: MCP server tool` in the approval prompt.
Strict `mcp__` prefix only.
- Revert the cargo-cult `{ autoApprove: true } as AgentToolContext` insertions
in agent-session-python-cleanup.test.ts and sdk-move-cwd.test.ts. The tests
create sessions without passing settings, so the wrapper falls through to
approvalMode "auto" automatically; the explicit flag was unnecessary and
the `as AgentToolContext` cast hid that autoApprove lives on
CustomToolContext, not AgentToolContext.
- Document in commands/launch.ts the dual --auto-approve declaration (oclif
Flags for --help, manual parseArgs for runtime) so a future rename catches
both call sites.
- Promote the subagent caveat in docs/approval-mode.md to a callout near the
top: anything `task` is asked to do runs unattended once the parent task
call is approved.
Verification:
- bun test packages/coding-agent/test/tools/approval.test.ts → 75 pass / 0 fail
(was 57; +18 cases covering new remote-exec patterns, chmod symbolic, tee
/etc, isMcp negative, and eval/task/ast_edit/browser/write payload previews)
- bun test packages/coding-agent/test/tools/approval-mode.test.ts → 7 pass /
0 fail (was 7; +1 case asserting extensionRunner is always constructed)
- bun tsc --noEmit -p packages/coding-agent → clean
- bun x biome check . → clean
- Windows EBUSY tempdir-cleanup noise in agent-session-python-cleanup and
sdk-move-cwd is pre-existing on this branch (already documented in the
PR body) and absent on Linux CI.
- approval: user 'tool: deny' now wins over critical-pattern override
(the override only tightens allow->prompt; it must never re-arm a denied tool).
- approval: rename hindsight policy keys to match registered tool names
(recall/retain/reflect, not hindsight_recall/hindsight_retain).
- approval: head+tail truncation for bash/ssh command prompts so a
destructive suffix buried after a long benign preamble stays visible.
- task/executor: force tools.approvalMode='auto' in createSubagentSettings
so subagents (which have no UI) cannot deadlock on per-tool prompts;
the parent's approval of the task call is the authorization.
- docs/approval-mode: rewrite so every example surfaces tools.approvalMode
and explains that tools.approval is ignored outside 'custom' mode.
Re-introduces the per-tool approval system from luzidd's commit 39124f3 (which
is no longer reachable from main) and improves it before re-landing.
What's restored:
- ApprovalPolicy (allow/deny/prompt) plus DEFAULT_APPROVAL_POLICIES.
- ACTION_EXCEPTIONS registry (LSP read-only, bash critical patterns).
- getApprovalPolicy() six-level resolution order.
- ExtensionToolWrapper.execute() gate before extension handlers.
- --auto-approve / --yolo CLI flag and tools.approval.<tool> user config.
- docs/approval-mode.md user guide.
What's improved over the original:
- Replaced unchecked 'as any' casts with typed unknown narrowing helpers.
- Validate userConfig values: invalid strings, numbers, etc. fall through to
the built-in default instead of being silently honoured (typo no longer
locks a tool out or grants implicit approval).
- Expanded CRITICAL_BASH_PATTERNS: chmod -R /, chown -R /, bash <(curl ...),
writes to /etc/passwd|shadow|sudoers, shutdown/reboot/halt/init 0,
kill -9 1, nc -e / nc -c reverse shells. Pattern shapes require a
command-position boundary so 'npm run reboot-tests' and 'echo "shutdown the
queue"' don't false-positive.
- Added DEBUG_READONLY_ACTIONS exception so DAP inspection actions (threads,
stack_trace, variables, scopes, read_memory, …) auto-allow while
execution-side actions (launch, attach, continue, evaluate, write_memory,
set_breakpoint, …) still prompt.
- formatApprovalPrompt: labels mcp__<server>__<tool> calls as MCP server
tools, surfaces ssh host + command, recognises the modern § hashline header
for edit, and truncates >240-char fields so a heredoc-sized body cannot
blow out the confirmation dialog.
- Test suite grown from 40 to 57 cases — new coverage for invalid user
config, the extended critical-bash patterns, benign-keyword negatives,
debug exceptions, MCP/ssh prompt formatting, and command truncation.
Verification:
- bun test packages/coding-agent/test/tools/approval.test.ts -> 57 pass
- bun x biome check . -> clean
- bun run check:ts across all 9 workspaces -> clean
- Dropped the `fileType: natives.FileType.File` restriction so glob searches can return directories as well as files.
- Updated the find tool prompt to document directory results and trailing-slash output.
- Added tests verifying directory matches are included and emitted with a trailing `/`.
- Updated `TempDirGuard` creation in grep tests to include PID and an atomic sequence, preventing temp path collisions.
- Removed the `branch` filter from GitHub action run queries so results are matched by `head_sha` only.
- Adjusted run-watch calls to the simplified `fetchRunsForCommit` interface without the branch argument.
- Replaced external watchdog timers with per-request SDK timeouts for first-event budget across OpenAI, Anthropic, and Azure providers.
- Keyed Python shared kernels by (sessionId, cwd) to prevent cross-directory state bleed.
- Deduplicated concurrent cold-start session acquisition for JS and Python executors.
- Moved `isOpenAIResponsesProgressEvent` to shared module and scoped display output routing per run for interleaved async cells.
- Removed deprecated MCP-specific type aliases and functions from tool-discovery module, consolidating to unified generic tool discovery API.
- Migrated session and SDK code to use generic filterBySource() and collectDiscoverableTools() instead of MCP-specific variants.
- Removed deprecated interface members including hasQueuedMessages(), FocusPane, AcpBuiltinCommandRuntime, and legacy settings methods.
- Updated test suites to use renamed generic discovery methods and removed back-compat test coverage for legacy MCP shapes.
- Added `cwd` and `env` optional parameters to kernel execution API for runtime working directory and environment variable control.
- Implemented runtime environment setup in Python runner with `_apply_request_runtime()` to apply cwd and env from request before code execution.
- Enhanced SIGINT handler management with `active_executions` counter and `_begin_exec_sigint()` / `_end_exec_sigint()` functions to prevent state mutation during concurrent execution.
- Changed `SearchRenderArgs.paths` parameter type from `string[]` to `string | string[]` to accept single string paths.
- Added comprehensive test coverage for kernel cwd updates, timeout interruption safety, and SystemExit handling in shared executor sessions.
- Updated type signature to show `paths` accepts `string | string[]` instead of only arrays.
- Clarified that single string paths are wrapped into a one-element list before resolution.
- Improved prompt instructions to explicitly show both string and array usage patterns.
- Removed per-session run queues from JS and Python backends, allowing async cells on the same session id to interleave.
- Introduced `getEvalSessionId` on ToolSession so subagents spawned via `task` inherit the parent's executor id and share JS VM and Python kernel state.
- Switched JS runtime state from module-level fields to AsyncLocalStorage so concurrent runs route output and tool calls to their own context.
- Changed Python runner to an asyncio event loop with per-request tasks and ContextVar-based run id tracking for concurrent execution.
- Added mtime-based module cache eviction to preserve singleton state across re-imports of unchanged local files.
- Replaced per-line hash anchors with file-level hash validation in hashline format, changing anchor syntax from LINE+HASH to bare LINE numbers.
- Simplified hashline line separator from pipe (|) to colon (:) and replaced replace operator (->) with colon, added delete operator (!) for explicit line deletion.
- Implemented file-read snapshot caching with multi-snapshot ring buffer per path and file-hash-based recovery to detect and recover from stale edits.
- Refactored hashline grammar, parser, and execution to support file-level hash binding, anchor-scoped validation, and structural bracket warnings for delete operations.
- Updated documentation and test fixtures to reflect new hashline syntax with file hashes, colon separators, and delete operator throughout.
- Removed vim edit mode and automatically map existing vim configurations to hashline mode.
- Deleted VimTool class, VimEngine implementation, and all vim-specific editing logic (2409 lines).
- Removed vim mode from EditMode union type, edit tool strategies, and configuration schemas.
- Deleted vim parser, command handler, buffer manager, and renderer modules.
- Updated documentation and tests to remove vim mode references and add deprecation mapping.
Refactor:
- session.ts: extract mapDebugpyMissingModule helper; replace the duplicated
inline check in launch/attach catch blocks. Add jsdoc on DapStartRequestFailure.settled
documenting per-call ownership and how throwPreferredDapStartError consumes it.
- path-utils.ts: replace the no-op keepOpaqueResourceUri branch with an
OPAQUE_RESOURCE_SCHEMES Set so the structure carries the intent. Functionally
equivalent; new opaque schemes become a one-line Set change.
Tests:
- dap-launch-failures: cover the debugpy stderr -> 'pip install debugpy'
rewrite for launch and attach, plus a negative case (non-debugpy adapter
with the substring in stderr is left untouched).
- dap-launch-failures: model the delayed-launch-failure case the new
settled-race in throwPreferredDapStartError defends against. FakeDapClient
gains optional launchErrorDelayMs/attachErrorDelayMs.
- dap-launch-failures (DebugTool): assert adapter:'debugpy' early-throw
surfaces 'python not found in PATH' on both launch and attach when
selectLaunchAdapter/selectAttachAdapter return null, and the unspecified-adapter
path still falls back to the generic 'No debugger adapter' error.
- find.ts: export validateFindPathInputs and pin the new backslash-escape
semantics (\, no longer trips the comma-joined heuristic) plus the
existing brace-expansion and rejection paths.
- patch.ts: cover the post-write verification error message. The user-facing
ToolError must contain the caller-supplied relative path and not the
absolute resolvedPath (which still lives in the structured context for
log correlation).
- split-internal-url-sel: reword two mcp:// test comments that described a
'peeler refuses' guard that doesn't exist; rename the tests to reflect the
actual opaque-scheme rule.
- browser tool's existing-tab re-nav defaults to waitUntil: 'load' (matching
new-tab path); identical acquireTab() calls no longer hang on dev servers
- patch tool error path uses caller-supplied relative path; absolute
resolvedPath stays in structured context only ($HOME no longer leaks to TUI)
- splitInternalUrlSel keeps mcp:// resource URIs opaque even when they end in
':raw' or '/:1-50' (McpProtocolHandler matches by verbatim URI)
- find tool: timeout signal honored by onMatch; partial results sorted by
mtime desc; backslash-escaped commas skipped in path-list validation
- DAP throwPreferredDapStartError waits up to 50ms for the underlying
launch/attach error instead of one microtask
- debug tool surfaces 'python missing' and 'pip install debugpy' diagnostics
separately when adapter: 'debugpy' is requested
- Deleted the `ValidationVerdict` type and `evaluateOutputAgainstSchema` API from the output schema validator.
- Updated `yield.ts` to bind `buildOutputValidator`'s error directly to `schemaError` during validator setup.
- Removed the obsolete evaluator tests and adjusted validation success fixture to match the raw summary input shape.
- Unified output schema construction and validation by adding buildOutputValidator and using it in YieldTool and task executor.
- Added MAX_SCHEMA_RETRIES so YieldTool now retries schema failures three times with hints before overriding.
- Updated failure handling to use shared summarizeValidationFailure and formatters for required-field reporting.
- Added tests for output-schema-validator and YieldTool covering malformed schemas and nested-array retry edge cases.
- packages/ai/test/issue-957-repro.test.ts now tests:
- refreshKimiToken applies the 5-minute server-side skew (Kimi-specific)
- AuthStorage refreshes kimi-code credentials inside its 60s skew window
- packages/ai/test/anthropic-stream-timeout.test.ts: raise the
streamFirstEventTimeoutMs from 10ms to 5000ms so slow CI scheduling
cannot fire the first-event watchdog before the mocked events arrive.
The test still exercises the (1ms) idle path it was written for.
fix(web): allow Parallel extract via PARALLEL_API_KEY env var without storage
The fetch tool and YouTube scraper previously gated the Parallel extract
branch behind `storage && findParallelApiKey(storage)`. With no
AgentStorage the env key was never consulted, so callers that ran
without a per-session storage (e.g. ReadTool sessions in unit tests, and
in practice any caller that has only an env API key) silently fell back
to raw-html / no-ytdlp paths.
- findCredential/findParallelApiKey now accept null or undefined storage
and rely solely on the env-first path when no storage is supplied.
- searchWithParallel/extractWithParallel mirror the same nullable shape.
- Drop the redundant `storage && ` guards in fetch.ts and youtube.ts;
the inner findParallelApiKey call already returns null when no
credential is available.
- Added OpenAI Codex and Gemini web search provider options with updated setup/auth descriptions.
- Updated Codex OAuth flow to refresh near-expiry tokens during web_search and persist the refreshed credentials.
- Plumbed AgentStorage through search orchestrator, scrapers, and fetch paths so providers share session credentials.
- Refactored web provider and credential helpers to accept caller-provided AgentStorage and resolve keys synchronously.
The report_finding tool's priority is exposed as a string enum
("P0"-"P3") for ergonomics, but the reviewer agent and every
custom review agent declare priority as `type: number` in their
JTD output schema. The cast at executor.ts:1473 lied about the
runtime shape, so the auto-injected `findings[].priority` flowed
through as strings and every yield with at least one finding was
rejected with `findings.0.priority: expected number, received string`,
forcing the run into the schema_violation exit path.
Added `toReviewFinding(details)` in tools/review.ts that maps the
priority enum to its numeric ordinal via the existing PRIORITY_INFO
table and use it at the boundary in executor.ts. Render paths still
see the original `ReportFindingDetails` shape (string priority)
through normalizeReportFindings, so display formatting is unaffected.
Fixes#1350
The JTD-to-JSON-Schema converter post-processed convertSchema's
output with normalizeMixedSchemaNode, which walked back into the
emitted JSON Schema looking for nested JTD forms. Inside a
properties block, user-defined property names whose keys happened
to collide with JTD keywords ('ref', 'elements', 'values',
'optionalProperties', 'discriminator') were misclassified as JTD
forms and re-rewritten - corrupting properties like { ref: { type:
'string' } } into { $ref: '#/$defs/[object Object]' } and breaking
the built-in explore agent's output validator with
schema_violation: files.0.ref: must not be present.
convertSchema is already fully recursive and emits pure JSON Schema,
so the post-walk is both unnecessary and unsafe. Drop it.
Fixes#1345
- Removed the exported formatBashFixupNotice helper from bash command fixup utilities.
- Removed BashTool's one-time bash-fixup notice tracking and stopped emitting those notices when fixups were applied.
Appended the stealth iframe to documentElement when document.head is not available during new-document evaluation.
Added regression coverage for the null-head bootstrap path.
Fixes#1267
Cleanup tail of 2817c582a — the search archive commit (78841798f) had
inadvertently included the redaction.ts import and wiring in
search.ts. That ad-hoc redactor was already reverted; this drops the
matching call sites so the file no longer references the deleted
module. SecretObfuscator (gated on `secrets.enabled`) is the supported
path for redaction.
Adds a focused redaction utility that targets well-known token shapes
(AWS, GitHub PATs/tokens, Slack, OpenAI-style sk-, JWT) plus a
key/value heuristic for env-style lines whose key contains SECRET /
TOKEN / PASSWORD / API_KEY / PRIVATE_KEY, plus Bearer/Basic Authorization
header values. Replacements are tagged `#REDACTED:<hint>#` so callers
can tell why each value was scrubbed.
Wired into read, search and ssh tool outputs. Each call site appends a
`[redacted N secret-like values]` footer when at least one value was
scrubbed so the model knows the output was modified.
Gated behind a new `tools.redactSecrets` boolean setting (default true).
Sandboxed tests that intentionally surface secret-shaped fixtures can
disable it via Settings.isolated({ "tools.redactSecrets": false }).
`read` accepts `archive.zip:member` selectors but `search` previously
ignored them, returning zero matches even when the member's text
contained the query. resolveArchiveSearchPaths now detects archive
selectors, opens the archive via the shared archive-reader, decodes
UTF-8 members into a scratch tmpdir, and rewrites match paths back to
the original selector before returning. Binary, non-UTF-8, missing
members and unreadable archives surface as a structured error or a
per-archive footer note. Scratch dir is cleaned up in finally.
The read renderer was emitting `LINE+HASH|content` for lines whose
content had been column-truncated for display, but `computeLineHash`
is content-only and recomputes against the disk line. The model copied
the displayed anchor, edit rejected it as mismatched, even though the
underlying file had not changed.
formatTextWithMode now accepts an optional truncatedLines set; in
hashLines mode those lines emit as `LINE|content` (no hash) so the
verifier never tries to recompute against truncated text. Multi-range
and single-range read paths both populate the set when truncateLine
flips wasTruncated.
The narrowed object-with-title schema added in e26a17f3f is no longer
necessary: the upstream constrained-sampling fix lets models emit
arbitrary props on a record now, so the opaque shape no longer hides
title from discovery. Plan-approval callers still pass extra.title and
the renderer/handler logic accept it unchanged.
The companion changes from e26a17f3f (resolve.md context enumeration,
runResolveInvocation apply-throw requeue) stay intact.
- The plan-approval gate required extra.title but the wire schema only
declared an opaque additionalProperties record so codex/gpt-5.x could
not discover the field. Schema now declares title with a description
while still allowing passthrough for future per-context keys.
- resolve.md replaces the truncated "Schema depends on context:" line
with the actual enumeration.
- runResolveInvocation wraps apply() in try/catch; a thrown apply (e.g.
ast_edit overlap) requeues the resolve directive so the model can
discard or fix-and-retry instead of losing the preview.
- todo-write.md adds an explicit note that tasks are referenced by
verbatim content text; the tool never emits task-N IDs.
- resolveTaskOrError rejects ^task-\d+$ inputs with a clarifying error.
- execute sets isError:true when any op failed.
- appendItems short-circuits on the first "already exists" error so the
call no longer applies the prefix of a doomed batch.
can1357