fix(agents): harden edit tool recovery#52516
Merged
Conversation
🔒 Aisle Security AnalysisWe found 2 potential security issue(s) in this PR:
1. 🟠 Edit tool error message leaks current file contents on exact-match mismatch
DescriptionThe edit-tool recovery wrapper appends up to 800 characters of the current file contents to an exception message when it detects an exact-match mismatch error.
This creates an unintended read/exfiltration primitive for any file that the edit tool can access (e.g., workspace RecommendationDo not include raw file contents in exception messages/tool errors. Safer alternatives:
Example: replace the hint with metadata only: function appendMismatchHint(error: Error, currentContent: string, absolutePath: string): Error {
const enhanced = new Error(
`${error.message}\n` +
`Hint: file_length=${currentContent.length} path=${absolutePath}`
);
enhanced.stack = error.stack;
return enhanced;
}If you must show content, implement robust secret redaction (e.g., detect common credential patterns) and ensure it is not logged or returned in tool results. 2. 🟠 Workspace boundary bypass: edit recovery wrapper reads arbitrary absolute/tilde paths before policy enforcement
DescriptionThe This introduces a workspace/path-policy bypass risk when
Vulnerable code: const absolutePath =
typeof pathParam === "string" ? resolveEditPath(options.root, pathParam) : undefined;
if (absolutePath && newText !== undefined) {
originalContent = await options.readFile(absolutePath);
}RecommendationApply the same workspace/root path policy to recovery reads as the underlying edit operations. Options:
const operations = createHostEditOperations(root, { workspaceOnly: true });
const base = createEditTool(root, { operations }) as AnyAgentTool;
const withRecovery = wrapEditToolWithRecovery(base, {
root,
readFile: async (absolutePath) => (await operations.readFile(absolutePath)).toString("utf8"),
});
Also consider avoiding full-file reads for mismatch hints (read only the first N bytes) to reduce accidental sensitive-data exposure and memory/DoS risk. Analyzed PR: #52516 at commit Last updated on: 2026-03-23T09:54:52Z Latest run failed. Keeping previous successful results. Trace ID: Last updated on: 2026-03-24T15:06:22Z |
openclaw-barnacle
bot
added
agents
Contributor
Greptile SummaryThis PR hardens the existing edit-tool exact-match recovery path by unifying host and sandbox edit tools under a single Key observations:
Confidence Score: 4/5
Prompt To Fix All With AIThis is a comment left during a code review.
Path: src/agents/pi-tools.host-edit.ts
Line: 171
Comment:
**`pathParam` typed as `string | undefined` passed to `string` parameter**
`pathParam` is destructured from `readEditToolParams` whose return type is `EditToolParams`, where `pathParam?: string` — so its type is `string | undefined`. `buildEditSuccessResult` expects a `string`. TypeScript's control-flow analysis does not narrow `pathParam` based on the `absolutePath` guard above, so `pnpm tsgo` (strict mode) will reject this.
```suggestion
return buildEditSuccessResult(pathParam!);
```
Alternatively, an explicit `typeof pathParam === "string"` check would avoid the non-null assertion while keeping the intent clear.
How can I resolve this? If you propose a fix, please make it concise.
---
This is a comment left during a code review.
Path: src/agents/pi-tools.host-edit.ts
Line: 72-86
Comment:
**Deletion edits can produce false-positive recovery when snapshot is unavailable**
When `newText` is `""` (deletion) and `originalContent` is unavailable (the pre-execute snapshot failed), `didEditLikelyApply` can incorrectly return `true`:
- Check 1 is skipped (no `originalContent`).
- Check 2 is skipped (`normalizedNew.length === 0`).
- Check 3: `withoutInsertedNewText === normalizedCurrent`; if `oldText` happened to be absent from the file before the write (e.g., the base threw a non-mismatch error before writing), `normalizedCurrent.includes(normalizedOld)` is `false` → returns `true`.
The existing "does not recover false success when the file never changed" test passes only because the pre-execute snapshot *succeeds* in that case — if it had failed, the test would flip.
The comment "Best-effort snapshot only; recovery should still proceed without it" is intentional, but it means deletion recovery without a snapshot is weaker than the PR description implies. A simple early-exit when `originalContent` is `undefined` and `newText.length === 0` (or a note in the function docblock) would make this explicit.
How can I resolve this? If you propose a fix, please make it concise.Reviews (1): Last reviewed commit: "fix: harden edit tool recovery" | Re-trigger Greptile |
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
src/agents/pi-tools.host-edit.ts
Outdated
| newText, | ||
| }) | ||
| ) { | ||
| return buildEditSuccessResult(pathParam); |
Contributor
There was a problem hiding this comment.
pathParam typed as string | undefined passed to string parameter
pathParam is destructured from readEditToolParams whose return type is EditToolParams, where pathParam?: string — so its type is string | undefined. buildEditSuccessResult expects a string. TypeScript's control-flow analysis does not narrow pathParam based on the absolutePath guard above, so pnpm tsgo (strict mode) will reject this.
Suggested change
| return buildEditSuccessResult(pathParam); | |
| return buildEditSuccessResult(pathParam!); |
Alternatively, an explicit typeof pathParam === "string" check would avoid the non-null assertion while keeping the intent clear.
Prompt To Fix With AI
This is a comment left during a code review.
Path: src/agents/pi-tools.host-edit.ts
Line: 171
Comment:
**`pathParam` typed as `string | undefined` passed to `string` parameter**
`pathParam` is destructured from `readEditToolParams` whose return type is `EditToolParams`, where `pathParam?: string` — so its type is `string | undefined`. `buildEditSuccessResult` expects a `string`. TypeScript's control-flow analysis does not narrow `pathParam` based on the `absolutePath` guard above, so `pnpm tsgo` (strict mode) will reject this.
```suggestion
return buildEditSuccessResult(pathParam!);
```
Alternatively, an explicit `typeof pathParam === "string"` check would avoid the non-null assertion while keeping the intent clear.
How can I resolve this? If you propose a fix, please make it concise.
Comment on lines
+72
to
+86
| if (normalizedOriginal !== undefined && normalizedOriginal === normalizedCurrent) { | ||
| return false; | ||
| } | ||
|
|
||
| if (normalizedNew.length > 0 && !normalizedCurrent.includes(normalizedNew)) { | ||
| return false; | ||
| } | ||
|
|
||
| const withoutInsertedNewText = | ||
| normalizedNew.length > 0 | ||
| ? removeExactOccurrences(normalizedCurrent, normalizedNew) | ||
| : normalizedCurrent; | ||
| if (normalizedOld && withoutInsertedNewText.includes(normalizedOld)) { | ||
| return false; | ||
| } |
Contributor
There was a problem hiding this comment.
When newText is "" (deletion) and originalContent is unavailable (the pre-execute snapshot failed), didEditLikelyApply can incorrectly return true:
- Check 1 is skipped (no
originalContent). - Check 2 is skipped (
normalizedNew.length === 0). - Check 3:
withoutInsertedNewText === normalizedCurrent; ifoldTexthappened to be absent from the file before the write (e.g., the base threw a non-mismatch error before writing),normalizedCurrent.includes(normalizedOld)isfalse→ returnstrue.
The existing "does not recover false success when the file never changed" test passes only because the pre-execute snapshot succeeds in that case — if it had failed, the test would flip.
The comment "Best-effort snapshot only; recovery should still proceed without it" is intentional, but it means deletion recovery without a snapshot is weaker than the PR description implies. A simple early-exit when originalContent is undefined and newText.length === 0 (or a note in the function docblock) would make this explicit.
Prompt To Fix With AI
This is a comment left during a code review.
Path: src/agents/pi-tools.host-edit.ts
Line: 72-86
Comment:
**Deletion edits can produce false-positive recovery when snapshot is unavailable**
When `newText` is `""` (deletion) and `originalContent` is unavailable (the pre-execute snapshot failed), `didEditLikelyApply` can incorrectly return `true`:
- Check 1 is skipped (no `originalContent`).
- Check 2 is skipped (`normalizedNew.length === 0`).
- Check 3: `withoutInsertedNewText === normalizedCurrent`; if `oldText` happened to be absent from the file before the write (e.g., the base threw a non-mismatch error before writing), `normalizedCurrent.includes(normalizedOld)` is `false` → returns `true`.
The existing "does not recover false success when the file never changed" test passes only because the pre-execute snapshot *succeeds* in that case — if it had failed, the test would flip.
The comment "Best-effort snapshot only; recovery should still proceed without it" is intentional, but it means deletion recovery without a snapshot is weaker than the PR description implies. A simple early-exit when `originalContent` is `undefined` and `newText.length === 0` (or a note in the function docblock) would make this explicit.
How can I resolve this? If you propose a fix, please make it concise.![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: ac93a63cd6
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
Comment on lines
+72
to
+74
| if (normalizedOriginal !== undefined && normalizedOriginal === normalizedCurrent) { | ||
| return false; | ||
| } |
There was a problem hiding this comment.
In didEditLikelyApply, the early equality guard compares originalContent and currentContent after normalizing line endings to LF. That means a successful edit that only changes CRLF↔LF is treated as "no change," so post-write recovery never triggers and the tool rethrows even though the file was updated. This regresses the wrapper’s goal for post-write throw recovery in newline-only edits; the equality short-circuit should use raw content (or a separate raw-difference check) before normalized matching.
Useful? React with 👍 / 👎.
mbelinky
force-pushed
the
codex/edit-recovery-hardening
branch
from
ca9fa7c to
23ede63
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: eca61398c5
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| const withRecovery = wrapHostEditToolWithPostWriteRecovery(base, root); | ||
| const withRecovery = wrapEditToolWithRecovery(base, { | ||
| root, | ||
| readFile: (absolutePath: string) => fs.readFile(absolutePath, "utf-8"), |
There was a problem hiding this comment.
When workspaceOnly is enabled, this host edit constructor now passes raw fs.readFile into wrapEditToolWithRecovery, but that wrapper performs pre/post-readback before or outside the guarded edit operations. In the direct createHostWorkspaceEditTool(..., { workspaceOnly: true }) path, those recovery reads bypass createHostEditOperations root enforcement, so an out-of-workspace path can still be read (and potentially surfaced via mismatch hints) even though the edit itself is supposed to be confined to the workspace.
Useful? React with 👍 / 👎.
mbelinky
force-pushed
the
codex/edit-recovery-hardening
branch
from
eca6139 to
1c1ec99
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 1c1ec99c98
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| hasStableTarget = | ||
| appendFingerprintAlias(parts, record, "newpath", ["newPath", "new_path"]) || hasStableTarget; | ||
| hasStableTarget = | ||
| appendFingerprintAlias(parts, record, "to", ["to", "target"]) || hasStableTarget; |
There was a problem hiding this comment.
This now prefers to over target when building mutation fingerprints, but message execution treats target as canonical and explicitly discards legacy to when both are present (normalizeMessageActionInput in src/infra/outbound/message-action-normalization.ts). If a failed call includes conflicting to/target, the stored fingerprint can point at the wrong destination and a later success on that to value can incorrectly clear an unrelated unresolved mutating error.
Useful? React with 👍 / 👎.
| if (alias in normalized && !(original in normalized)) { | ||
| normalized[original] = normalized[alias]; | ||
| } | ||
| delete normalized[alias]; |
There was a problem hiding this comment.
Alias keys are deleted unconditionally even when the canonical key already exists but is invalid (for example empty or non-string), so a payload like { path: "", file: "notes.md" } loses the usable alias before validation and is rejected as missing required path. This undermines the alias-compatibility hardening in this change for mixed-key model outputs; alias values should only be dropped after a usable canonical value is confirmed.
Useful? React with 👍 / 👎.
Contributor
Author
|
Merged via squash.
Thanks @mbelinky! |
This was referenced Mar 24, 2026
tiagonix
pushed a commit
to tiagonix/openclaw
that referenced
this pull request
Mar 24, 2026
Merged via squash. Prepared head SHA: e23bde8 Co-authored-by: mbelinky <[email protected]> Co-authored-by: mbelinky <[email protected]> Reviewed-by: @mbelinky
siofra-seksbot
added a commit
to TheBotsters/botster-ego
that referenced
this pull request
Mar 25, 2026
* Formatting fixes and remove trailing dash acceptance * Remove lower casing -- preserving prior behavior * fix: preserve legacy clawhub skill updates (openclaw#53206) (thanks @drobison00) * feat(csp): support inline script hashes in Control UI CSP (openclaw#53307) thanks @BunsDev Co-authored-by: BunsDev <[email protected]> Co-authored-by: Nova <[email protected]> * refactor: separate exec policy and execution targets * test: print failed test lane output tails * fix(cron): make --tz work with --at for one-shot jobs Previously, `--at` with an offset-less ISO datetime (e.g. `2026-03-23T23:00:00`) was always interpreted as UTC, even when `--tz` was provided. This caused one-shot jobs to fire at the wrong time. Changes: - `parseAt()` now accepts an optional `tz` parameter - When `--tz` is provided with `--at`, offset-less datetimes are interpreted in that IANA timezone using Intl.DateTimeFormat - Datetimes with explicit offsets (e.g. `+01:00`, `Z`) are unaffected - Removed the guard in cron-edit that blocked `--tz` with `--at` - Updated `--at` help text to mention `--tz` support - Added 2 tests verifying timezone resolution and offset preservation * fix: land cron tz one-shot handling and prerelease config warnings (openclaw#53224) (thanks @RolfHegr) * fix: clean changelog merge duplication (openclaw#53224) (thanks @RolfHegr) * test: isolate line jiti runtime smoke * refactor: harden extension runtime-api seams * tests: improve boundary audit coverage and safety (openclaw#53080) * tools: extend seam audit inventory * tools: tighten seam audit heuristics * tools: refine seam test matching * tools: refine seam audit review heuristics * style: format seam audit script * tools: widen seam audit matcher coverage * tools: harden seam audit coverage * tools: tighten boundary audit matchers * tools: ignore mocked import matches in boundary audit * test: include native command reply seams in audit * fix: command auth SecretRef resolution (openclaw#52791) (thanks @Lukavyi) * fix(command-auth): handle unresolved SecretRef in resolveAllowFrom * fix(command-auth): fall back to config allowlists * fix(command-auth): avoid duplicate resolution fallback * fix(command-auth): fail closed on invalid allowlists * fix(command-auth): isolate fallback resolution errors * fix: record command auth SecretRef landing notes (openclaw#52791) (thanks @Lukavyi) --------- Co-authored-by: Ayaan Zaidi <[email protected]> * refactor: extract cron schedule and test runner helpers * fix: populate currentThreadTs in threading tool context fallback for Telegram DM topics (openclaw#52217) When a channel plugin lacks a custom buildToolContext (e.g. Telegram), the fallback path in buildThreadingToolContext did not set currentThreadTs from the inbound MessageThreadId. This caused resolveTelegramAutoThreadId to return undefined, so message tool sends without explicit threadId would route to the main chat instead of the originating DM topic. Fixes openclaw#52217 * fix: unblock runtime-api smoke checks * refactor: split tracked ClawHub update flows * build: prepare 2026.3.23-2 * fix: preserve command auth resolution errors on empty inferred allowlists * docs: refresh plugin-sdk api baseline * test: harden linux runtime smoke guards * fix(runtime): anchor bundled plugin npm staging to active node * tests: cron coverage and NO_REPLY delivery fixes (openclaw#53366) * tools: extend seam audit inventory * tools: audit cron seam coverage gaps * test: add cron seam coverage tests * fix: avoid marking NO_REPLY cron deliveries as delivered * fix: clean up delete-after-run NO_REPLY cron sessions * fix: verify global npm correction installs * build: prepare 2026.3.24 * docs: update mac release automation guidance * fix: fail closed when provider inference drops errored allowlists * fix: reject nonexistent zoned cron at-times * fix: hash inline scripts with data-src attributes * ci: balance shards and reuse pr artifacts * refactor: simplify provider inference and zoned parsing helpers * fix: unify live model auth gating * tests: add boundary coverage for media delivery (openclaw#53361) * tests: add boundary coverage for media delivery * tests: isolate telegram outbound adapter transport * tests: harden telegram webhook certificate assertion * tests: fix guardrail false positives on rebased branch * msteams: extract structured quote/reply context (openclaw#51647) * msteams: extract structured quote/reply context from Teams HTML attachments * msteams: address PR openclaw#51647 review feedback * msteams: add message edit and delete support (openclaw#49925) - Add edit/delete action handlers with toolContext.currentChannelId fallback for in-thread edits/deletes without explicit target - Add editMessageMSTeams/deleteMessageMSTeams to channel runtime - Add updateActivity/deleteActivity to SendContext and MSTeamsTurnContext - Extend content param with text/content/message fallback chain - Update test mocks for new SendContext shape Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]> * fix(doctor): honor --fix in non-interactive mode Ensure repair-mode doctor prompts auto-accept recommended fixes even when running non-interactively, while still requiring --force for aggressive rewrites. This restores the expected behavior for upgrade/doctor flows that rely on 'openclaw doctor --fix --non-interactive' to repair stale gateway service configuration such as entrypoint drift after global updates. Co-authored-by: Copilot <[email protected]> * Preserve no-restart during update doctor fixes Co-authored-by: Copilot <[email protected]> * fix(doctor): skip service config repairs during updates Co-authored-by: Copilot <[email protected]> * fix: add config clobber forensics * fix(ui): resolve model provider from catalog instead of stale session default When the server returns a bare model name (e.g. "deepseek-chat") with a session-level modelProvider (e.g. "zai"), the UI blindly prepends the provider — producing "zai/deepseek-chat" instead of the correct "deepseek/deepseek-chat". This causes "model not allowed" errors when switching between models from different providers. Root cause: resolveModelOverrideValue() and resolveDefaultModelValue() in app-render.helpers.ts, plus the /model slash command handler in slash-command-executor.ts, all call resolveServerChatModelValue() which trusts the session's default provider. The session provider reflects the PREVIOUS model, not the newly selected one. Fix: for bare model names, create a raw ChatModelOverride and resolve through normalizeChatModelOverrideValue() which looks up the correct provider from the model catalog. Falls back to server-provided provider only if the catalog lookup fails. All 3 call sites are fixed. Closes openclaw#53031 Co-Authored-By: Claude Opus 4.6 <[email protected]> Signed-off-by: HCL <[email protected]> * style(ui): polish agent file preview and usage popovers (openclaw#53382) * feat: make workspace links clickable in agent context card and files list Updated the agent context card and files list to render workspace names as clickable links, allowing users to easily access the corresponding workspace files. This enhances usability by providing direct navigation to the workspace location. * style(ui): polish markdown preview dialog * style(ui): reduce markdown preview list indentation * style(ui): update markdown preview dialog width and alignment * fix(ui): open usage filter popovers toward the right * style(ui): adjust positioning of usage filter and export popovers * style(ui): update sidebar footer padding and modify usage header z-index * style(ui): adjust positioning of usage filter popover to the left and export popover to the right * style(ui): simplify workspace link rendering in agent context card * UI: make workspace paths interactive buttons or plain text Agent Context card workspace (Channels/Cron panels): replace non-interactive <div> with a real <button> wired to onSelectPanel('files'), matching the Overview panel pattern. Core Files footer workspace: drop workspace-link class since the user is already on the Files panel — keep as plain text. * fix(agents): suppress heartbeat prompt for cron-triggered embedded runs Prevent cron-triggered embedded runs from inheriting the default heartbeat prompt so non-cron session targets stop reading HEARTBEAT.md and polluting scheduled turns. Made-with: Cursor * test(agents): cover additional heartbeat prompt triggers Document that default-agent heartbeat prompt injection still applies to memory-triggered and triggerless runs while cron remains excluded. Made-with: Cursor * fix: land cron heartbeat prompt suppression (openclaw#53152) (thanks @Protocol-zero-0) * msteams: implement Teams AI agent UX best practices (openclaw#51808) Migrates the Teams extension from @microsoft/agents-hosting to the official Teams SDK (@microsoft/teams.apps + @microsoft/teams.api) and implements Microsoft's AI UX best practices for Teams agents. - AI-generated label on all bot messages (Teams native badge + thumbs up/down) - Streaming responses in 1:1 chats via Teams streaminfo protocol - Welcome card with configurable prompt starters on bot install - Feedback with reflective learning (negative feedback triggers background reflection) - Typing indicators for personal + group chats (disabled for channels) - Informative status updates (progress bar while LLM processes) - JWT validation via Teams SDK createServiceTokenValidator - User-Agent: teams.ts[apps]/<sdk-version> OpenClaw/<version> on outbound requests - Fix copy-pasted image downloads (smba.trafficmanager.net auth allowlist) - Pre-parse auth gate (reject unauthenticated requests before body parsing) - Reflection dispatcher lifecycle fix (prevent leaked dispatchers) - Colon-safe session filenames (Windows compatibility) - Cooldown cache eviction (prevent unbounded memory growth) Closes openclaw#51806 * refactor: tighten embedded prompt and sidecar guards * test: audit subagent seam coverage inventory * test: add exact-stem subagent seam tests * refactor: clarify doctor repair flow * fix(plugins): make Matrix recovery paths tolerate stale plugin config (openclaw#52899) * fix(plugins): address review feedback for Matrix recovery paths (openclaw#52899) 1. Narrow loadConfigForInstall() to catch only INVALID_CONFIG errors, letting real failures (fs permission, OOM) propagate. 2. Assert allow array is properly cleaned in stale-cleanup test. 3. Add comment clarifying version-resolution is already addressed via the shared VERSION constant. 4. Run cleanStaleMatrixPluginConfig() during install so persistPluginInstall() → writeConfigFile() does not fail validation on stale Matrix load paths. * fix(plugins): address review feedback for Matrix recovery paths (openclaw#52899) * fix: fetch model catalog for slash command updates * fix: restore teams sdk adapter contracts * fix: keep slash command model qualification on rebase * fix: clear production dependency advisories * fix: delete subagent runs after announce give-up * refactor: polish trigger and manifest seams * refactor(ui): extract chat model resolution state * fix(feishu): preserve docx block tree order (openclaw#40524) Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm vitest run extensions/feishu/src/docx.test.ts Co-authored-by: Tao Xie <[email protected]> * fix: stabilize matrix and teams ci assertions * fix: preserve subagent ended hooks until runtime init * test: prune low-signal live model sweeps * test: harden parallels smoke harness * fix: preserve direct subagent dispatch failures on abort * fix: report dropped subagent announce queue deliveries * fix: unblock live harness provider discovery * fix: finalize resumed subagent cleanup give-ups * refactor: centralize plugin install config policy * fix: format subagent registry test * fix: finalize deferred subagent expiry cleanup * fix(tui): preserve user message during slow model responses (openclaw#53115) When a local run ends with an empty final event while another run is active, skip history reload to prevent clearing the user's pending message from the chat log. This fixes the 'message disappears' issue with slow models like Ollama. * fix: preserve deferred TUI history sync (openclaw#53130) (thanks @joelnishanth) * test: sync app chat model override expectation * feat(ui): Control UI polish — skills revamp, markdown preview, agent workspace, macOS config tree (openclaw#53411) thanks @BunsDev Co-authored-by: BunsDev <[email protected]> Co-authored-by: Nova <[email protected]> * fix(security): resolve Aisle findings — skill installer validation, terminal sanitization, URL scheme allowlisting (openclaw#53471) thanks @BunsDev Co-authored-by: BunsDev <[email protected]> Co-authored-by: Nova <[email protected]> * fix: widen installer regex allowlists and deduplicate safeExternalHref calls - SAFE_GO_MODULE: allow uppercase in module paths (A-Z) - SAFE_BREW_FORMULA: allow @ for versioned formulas ([email protected]) - SAFE_UV_PACKAGE: allow extras [standard] and equality pins == - Cache safeExternalHref result in skills detail API key section * docs: update CONTRIBUTING.md * test: continue vitest threads migration * test: continue vitest threads migration * test: harden threaded shared-worker suites * test: harden threaded channel follow-ups * test: defer slack bolt interop for helper-only suites * fix(agents): harden edit tool recovery (openclaw#52516) Merged via squash. Prepared head SHA: e23bde8 Co-authored-by: mbelinky <[email protected]> Co-authored-by: mbelinky <[email protected]> Reviewed-by: @mbelinky * fix(docs): correct json55 typo to json5 in IRC channel docs (openclaw#50831) (openclaw#50842) Merged via squash. Prepared head SHA: 0f743bf Co-authored-by: Hollychou924 <[email protected]> Co-authored-by: altaywtf <[email protected]> Reviewed-by: @altaywtf * fix(secrets): prevent unresolved SecretRef from crashing embedded agent runs Root cause: Telegram channel monitor captures config at startup before secrets are resolved and passes it as configOverride into the reply pipeline. Since getReplyFromConfig() uses configOverride directly (skipping loadConfig() which reads the resolved runtime snapshot), the unresolved SecretRef objects propagate into FollowupRun.run.config and crash runEmbeddedPiAgent(). Fix (defense in depth): - get-reply.ts: detect unresolved SecretRefs in configOverride and fall back to loadConfig() which returns the resolved runtime snapshot - message-tool.ts: try-catch around schema/description building at tool creation time so channel discovery errors don't crash the agent - message-tool.ts: detect unresolved SecretRefs in pre-bound config at tool execution time and fall back to gateway secret resolution Fixes: openclaw#45838 * fix: merge explicit reply config overrides onto fresh config * fix: clean up failed non-thread subagent spawns * fix: initialize plugins before killed subagent hooks * fix: report qmd status counts from real qmd manager (openclaw#53683) (thanks @neeravmakwana) * fix(memory): report qmd status counts from index * fix(memory): reuse full qmd manager for status * fix(memory): harden qmd status manager lifecycle * fix: ci * fix: finalize killed delete-mode subagent cleanup * fix: clean up attachments for killed subagent runs * feat(cli): support targeting running containerized openclaw instances (openclaw#52651) Signed-off-by: sallyom <[email protected]> * fix: ci * Telegram: recover General topic bindings (openclaw#53699) Merged via squash. Prepared head SHA: 546f0c8 Co-authored-by: huntharo <[email protected]> Co-authored-by: huntharo <[email protected]> Reviewed-by: @huntharo * fix: clean up attachments for released subagent runs * fix(ci): do not cancel in-progress main runs * fix: clean up attachments for orphaned subagent runs * test: speed up discord extension suites * test: speed up slack extension suites * test: speed up telegram extension suites * test: speed up whatsapp and shared test suites * fix(ci): do not cancel in-progress bun runs on main * fix: clean up attachments when replacing subagent runs * feat(discord): add autoThreadName 'generated' strategy (openclaw#43366) * feat(discord): add autoThreadName 'generated' strategy Adds async thread title generation for auto-created threads: - autoThread: boolean - enables/disables auto-threading - autoThreadName: 'message' | 'generated' - naming strategy - 'generated' uses LLM to create concise 3-6 word titles - Includes channel name/description context for better titles - 10s timeout with graceful fallback * Discord: support non-key auth for generated thread titles * Discord: skip fallback auto-thread rename * Discord: normalize generated thread title first content line * Discord: split thread title generation helpers * Discord: tidy thread title generation constants and order * Discord: use runtime fallback model resolution for thread titles * Discord: resolve thread-title model aliases * Discord: fallback thread-title model selection to runtime defaults * Agents: centralize simple completion runtime * fix(discord): pass apiKey to complete() for thread title generation The setRuntimeApiKey approach only works for full agent runs that use authStorage.getApiKey(). The pi-ai complete() function expects apiKey directly in options or falls back to env vars — it doesn't read from authStorage.runtimeOverrides. Fixes thread title generation for Claude/Anthropic users. * fix(agents): return exchanged Copilot token from prepareSimpleCompletionModel The recent thread-title fix (3346ba6) passes prepared.auth.apiKey to complete(). For github-copilot, this was still the raw GitHub token rather than the exchanged runtime token, causing auth failures. Now setRuntimeApiKeyForCompletion returns the resolved token and prepareSimpleCompletionModel includes it in auth.apiKey, so both the authStorage path and direct apiKey pass-through work correctly. * fix(agents): catch auth lookup exceptions in completion model prep getApiKeyForModel can throw for credential issues (missing profile, etc). Wrap in try/catch to return { error } for fail-soft handling rather than propagating rejected promises to callers like thread title generation. * Discord: strip markdown wrappers from generated thread titles * Discord/agents: align thread-title model and local no-auth completion headers * Tests: import fresh modules for mocked thread-title/simple-completion suites * Agents: apply exchanged Copilot baseUrl in simple completions * Discord: route thread runtime imports through plugin SDK * Lockfile: add Discord pi-ai runtime dependency * Lockfile: regenerate Discord pi-ai runtime dependency entries * Agents: use published Copilot token runtime module * Discord: refresh config baseline and lockfile * Tests: split extension runs by isolation * Discord: add changelog for generated thread titles (openclaw#43366) (thanks @davidguttman) --------- Co-authored-by: Onur Solmaz <[email protected]> Co-authored-by: Onur Solmaz <[email protected]> * add missing autoArchiveDuration to DiscordGuildChannelConfig type (openclaw#43427) * add missing autoArchiveDuration to DiscordGuildChannelConfig type The autoArchiveDuration field is present in the Zod schema (DiscordGuildChannelSchema) and actively used at runtime in threading.ts and allow-list.ts, but was missing from the canonical TypeScript type definition. Add autoArchiveDuration to DiscordGuildChannelConfig to align the type with the schema and runtime usage. * Discord: add changelog for config type fix (openclaw#43427) (thanks @davidguttman) --------- Co-authored-by: Onur Solmaz <[email protected]> * refactor: dedupe test and script helpers * test: speed up discord extension suites * test: speed up slack extension suites * test: speed up telegram extension suites * test: speed up signal and whatsapp extension suites * fix(discord): avoid bundling pi-ai runtime deps * fix(lockfile): sync discord dependency removal * test: speed up discord slack telegram suites * test: speed up whatsapp and signal suites * test: speed up google and twitch suites * test: speed up core unit suites * fix: preserve cleanup hooks after subagent register failure * fix: preserve session cleanup hooks after subagent announce * Feishu: avoid CLI startup failure on unresolved SecretRef * fix(doctor): add missing baseUrl and models when migrating nano-banana apiKey to google provider The legacy nano-banana-pro skill migration moves the Gemini API key to models.providers.google.apiKey but does not populate the required baseUrl and models fields on the provider entry. When the google provider object is freshly created (no pre-existing config), the resulting config fails Zod validation on write: Config validation failed: models.providers.google.baseUrl: Invalid input: expected string, received undefined Fix: default baseUrl to 'https://generativelanguage.googleapis.com' and models to [] when they are not already set, matching the defaults used elsewhere in the codebase (embeddings-gemini, pdf-native-providers). Fixes the 'doctor --fix' crash for users who only have a legacy nano-banana-pro skill entry and no existing models.providers.google. * fix: use v1beta for migrated google nano banana provider (openclaw#53757) (thanks @mahopan) * docs: add changelog for PR openclaw#53675 (thanks @hpt) * fix(msteams): harden feedback reflection follow-ups * test: stabilize preaction process title assertion (openclaw#53808) Regeneration-Prompt: | Current origin/main fails src/cli/program/preaction.test.ts because the test asserts on process.title directly inside Vitest, where that runtime interaction is not stable enough to observe the write reliably. Keep the production preaction behavior unchanged. Make the test verify that the hook assigns the expected title by wrapping process.title with a local getter/setter during each test and restoring the original descriptor afterward so other tests keep the real process object behavior. * fix(auth): protect fresher codex reauth state - invalidate cached Codex CLI credentials when auth.json changes within the TTL window - skip external CLI sync when the stored Codex OAuth credential is newer - cover both behaviors with focused regression tests Refs openclaw#53466 Co-authored-by: Copilot <[email protected]> * fix: return structured errors for subagent control send failures * refactor: centralize google API base URL handling * refactor(msteams): split reply and reflection helpers * refactor(auth): unify external CLI credential sync * refactor: split feishu runtime and inspect secret resolution * test(memory): clear browser and plugin caches between cases * fix(types): add workspace module shims * fix: avoid duplicate orphaned subagent resumes * test(memory): enable lower-interval heap snapshots * fix: audit clobbered config reads * fix(whatsapp): filter fromMe messages in groups to prevent infinite loop (openclaw#53386) * fix: suppress only recent whatsapp group echoes (openclaw#53624) (thanks @w-sss) * test: speed up slack and telegram suites * test: speed up cli and model command suites * test: speed up command runtime suites * test: speed up backup and doctor suites * fix(memory): avoid caching status-only managers * fix: stabilize logging config imports * fix(slack): improve interactive reply parity (openclaw#53389) * fix(slack): improve interactive reply parity * fix(slack): isolate reply interactions from plugins * docs(changelog): note slack interactive parity fixes * fix(slack): preserve preview text for local agent replies * fix(agent): preserve directive text in local previews * test: preserve child_process exports in restart bun mock * fix(memory): avoid caching qmd status managers * test: speed up browser and gateway suites * test: speed up media fetch suite * fix(acp): deliver final result text as fallback when no blocks routed - Check routedCounts.final to detect prior delivery - Skip fallback for ttsMode='all' to avoid duplicate TTS processing - Use delivery.deliver for proper routing in cross-provider turns - Fixes openclaw#46814 where ACP child run results were not delivered * fix: tighten ACP final fallback semantics (openclaw#53692) (thanks @w-sss) * fix: unify pi runner usage snapshot fallback * refactor: isolate ACP final delivery flow * fix(ci): stop dropping pending main workflow runs * test(memory): isolate new unit hotspot files * test(memory): isolate browser remote-tab hotspot * test(memory): isolate plugin-core hotspot * test(memory): isolate telegram bot hotspot * fix: continue subagent kill after session store write failures * test(memory): isolate telegram fetch hotspot * test: speed up plugin-sdk and cron suites * test: speed up browser suites * test(memory): isolate telegram monitor hotspot * test(memory): isolate slack action-runtime hotspot * test(memory): recycle shared channels batches * fix: fail closed when subagent steer remap fails * Providers: fix kimi-coding thinking normalization * Providers: fix kimi fallback normalization * Plugins: resolve sdk aliases from the running CLI * Plugins: trust only startup cli sdk roots * Plugins: sanitize sdk export subpaths * Webchat: handle bare /compact as session compaction * Chat UI: tighten compact transport handling * Chat UI: guard compact retries * fix: ignore stale subagent steer targets * fix(discord): notify user on discord when inbound worker times out (openclaw#53823) * fix(discord): notify user on discord when inbound worker times out. * fix(discord): notify user on discord when inbound worker times out. * Discord: await timeout fallback reply * Discord: add changelog for timeout reply fix (openclaw#53823) (thanks @Kimbo7870) --------- Co-authored-by: VioGarden <[email protected]> Co-authored-by: Onur Solmaz <[email protected]> * refactor(channels): route registry lookups through runtime * refactor(plugins): make runtime registry lazy * refactor(plugins): make hook runner global lazy * refactor(plugins): make command registry lazy * fix: allow compact retry after failed session compaction (openclaw#53875) * refactor(gateway): make plugin fallback state lazy * refactor(plugins): make interactive state lazy * fix(memory): align status manager concurrency test * fix(runtime): stabilize dist runtime artifacts (openclaw#53855) * fix(build): stabilize lazy runtime entry paths * fix(runtime): harden bundled plugin npm staging * docs(changelog): note runtime artifact fixes * fix(runtime): stop trusting npm_execpath * fix(runtime): harden Windows npm staging * fix(runtime): add safe Windows npm fallback * ci: start required checks earlier (openclaw#53844) * ci: start required checks earlier * ci: restore pnpm in security-fast * ci: skip docs-only payloads in early check jobs * ci: harden untrusted pull request execution * ci: pin gradle setup action * ci: normalize pull request concurrency cancellation * ci: remove duplicate early-lane setup * ci: keep install-smoke push runs unique * fix: unblock supervisor and memory gate failures * test: stabilize low-profile parallel gate * refactor(core): make event and queue state lazy * fix(ci): refresh plugin sdk baseline and formatting * chore: refresh plugin sdk api baseline * fix: ignore stale subagent kill targets * perf(plugins): scope web search plugin loads * fix: ignore stale subagent send targets * fix: validate agent workspace paths before writing identity files (openclaw#53882) * fix: validate agent workspace paths before writing identity files * Feedback updates and formatting fixes * refactor: dedupe tests and harden suite isolation * test: fix manifest registry fixture typing * fix: ignore stale bulk subagent kill targets * fix(cli): precompute bare root help startup path * fix(test): stabilize npm runner path assertion * test(gateway): align safe open error code * test: speed up targeted unit suites * fix: prefer current subagent targets over stale rows * fix(ci): use target-platform npm path semantics * Adjust CLI backend environment handling before spawn (openclaw#53921) security(agents): sanitize CLI backend env overrides before spawn * fix: surface finished subagent send targets * perf(memory): avoid eager provider init on empty search * fix(test): satisfy cli backend config typing * fix: let subagent kill cascade through ended parents * perf(sqlite): use existence probes for empty memory search * fix: allow follow-up sends to finished subagents * fix: steer ended subagent orchestrators with live descendants * test: speed up browser pw-tools-core suites * test: speed up memory and secrets suites * fix(ci): align lazy memory provider tests * fix(test): stabilize memory vector dedupe assertion * fix(test): isolate github copilot token imports * fix: keep active-descendant subagents visible in reply status * refactor: dedupe helpers and source seams * test: fix rebase gate regressions * Adjust Feishu webhook request body limits (openclaw#53933) * fix: dedupe stale subagent rows in reply views * ci: batch shared extensions test lane * fix: report deduped subagent totals * fix: dedupe verbose subagent status counts * fix: align /agents ids with subagent targets * refactor: dedupe test helpers and harnesses * perf(memory): builtin sqlite hot-path follow-ups (openclaw#53939) * chore(perf): start builtin sqlite hotpath workstream * perf(memory): reuse sqlite statements during sync * perf(memory): snapshot file state during sync * perf(memory): consolidate status sqlite reads * docs(changelog): note builtin sqlite perf work * perf(memory): avoid session table scans on targeted sync * test: speed up memory provider suites * test: speed up slack monitor suites * test: speed up discord channel suites * test: speed up telegram and whatsapp suites * ci: increase test shard fanout * fix: clean up matrix /agents binding labels * fix: dedupe active child session counts * fix: dedupe restarted descendant session counts * fix: blcok non-owner authorized senders from chaning /send policy (openclaw#53994) * fix(slack): trim DM reply overhead and restore Codex auto transport (openclaw#53957) * perf(slack): instrument runtime and trim DM overhead * perf(slack): lazy-init draft previews * perf(slack): add turn summary diagnostics * perf(core): trim repeated runtime setup noise * perf(core): preselect default web search providers * perf(agent): restore OpenAI auto transport defaults * refactor(slack): drop temporary perf wiring * fix(slack): address follow-up review notes * fix(security): tighten slack and runtime defaults * style(web-search): fix import ordering * style(agent): remove useless spread fallback * docs(changelog): note slack runtime hardening * test: speed up discord monitor suites * test: speed up cli and command suites * test: speed up slack monitor suites * fix: ignore stale rows in subagent activity checks * fix: prefer latest subagent rows for session control * fix: ignore stale rows in subagent admin kill * fix: dedupe stale child completion announces * fix: ignore stale rows in subagent steer * fix: cascade bulk subagent kills past stale rows * fix: address FootGun's PR #8 review — regenerate metadata + fix Zulip imports 1. Regenerated bundled-plugin-metadata.generated.ts (stale after upstream merge) 2. Fixed Zulip extension monolithic plugin-sdk imports: - OpenClawPluginApi → openclaw/plugin-sdk/plugin-entry - emptyPluginConfigSchema, PluginRuntime, OpenClawConfig → openclaw/plugin-sdk/core - ChannelAccountSnapshot inline imports → openclaw/plugin-sdk/zulip 3. Added ChannelAccountSnapshot re-export to src/plugin-sdk/zulip.ts --------- Signed-off-by: HCL <[email protected]> Signed-off-by: sallyom <[email protected]> Co-authored-by: Devin Robison <[email protected]> Co-authored-by: Peter Steinberger <[email protected]> Co-authored-by: Val Alexander <[email protected]> Co-authored-by: BunsDev <[email protected]> Co-authored-by: Nova <[email protected]> Co-authored-by: Rolfy <[email protected]> Co-authored-by: Tak Hoffman <[email protected]> Co-authored-by: Taras Lukavyi <[email protected]> Co-authored-by: Ayaan Zaidi <[email protected]> Co-authored-by: Vincent Koc <[email protected]> Co-authored-by: sudie-codes <[email protected]> Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]> Co-authored-by: giulio-leone <[email protected]> Co-authored-by: Copilot <[email protected]> Co-authored-by: HCL <[email protected]> Co-authored-by: Protocol-zero-0 <[email protected]> Co-authored-by: Sid Uppal <[email protected]> Co-authored-by: Catalin Lupuleti <[email protected]> Co-authored-by: Tao Xie <[email protected]> Co-authored-by: Tao Xie <[email protected]> Co-authored-by: joelnishanth <[email protected]> Co-authored-by: Mariano <[email protected]> Co-authored-by: HollyChou <[email protected]> Co-authored-by: altaywtf <[email protected]> Co-authored-by: Neerav Makwana <[email protected]> Co-authored-by: Sally O'Malley <[email protected]> Co-authored-by: Harold Hunt <[email protected]> Co-authored-by: huntharo <[email protected]> Co-authored-by: David Guttman <[email protected]> Co-authored-by: Onur Solmaz <[email protected]> Co-authored-by: Onur Solmaz <[email protected]> Co-authored-by: Han Pingtian <[email protected]> Co-authored-by: Maho Pan <[email protected]> Co-authored-by: Josh Lehman <[email protected]> Co-authored-by: w-sss <[email protected]> Co-authored-by: scoootscooob <[email protected]> Co-authored-by: Bob <[email protected]> Co-authored-by: VioGarden <[email protected]> Co-authored-by: scoootscooob <[email protected]> Co-authored-by: Devin Robison <[email protected]>
netandreus
pushed a commit
to netandreus/openclaw
that referenced
this pull request
Mar 25, 2026
Merged via squash. Prepared head SHA: e23bde8 Co-authored-by: mbelinky <[email protected]> Co-authored-by: mbelinky <[email protected]> Reviewed-by: @mbelinky
npmisantosh
pushed a commit
to npmisantosh/openclaw
that referenced
this pull request
Mar 25, 2026
Merged via squash. Prepared head SHA: e23bde8 Co-authored-by: mbelinky <[email protected]> Co-authored-by: mbelinky <[email protected]> Reviewed-by: @mbelinky
godlin-gh
pushed a commit
to YouMindInc/openclaw
that referenced
this pull request
Mar 27, 2026
Merged via squash. Prepared head SHA: e23bde8 Co-authored-by: mbelinky <[email protected]> Co-authored-by: mbelinky <[email protected]> Reviewed-by: @mbelinky
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This PR cherry-picks the narrow, low-risk parts of the open edit-tool backlog into one cohesive fix.
It does not change
editsemantics, add fuzzy matching, add patch mode, or touch sandbox write-path integrity logic.Instead it hardens the existing exact-match flow in one shared recovery wrapper used by both host and sandbox edit tools, and extends parameter normalization for common alias spellings of the same exact-match operation.
oldTextis a substring ofnewTextnewTextis the empty stringisError: falsefile,filePath,old_text,new_text,oldString, andnewStringWhy this scope
The broader requests in #42221 and #44210 are real, but folding fuzzy matching, anchor mode, retry logic, or patch mode into the existing
edittool would make this much riskier.This PR stays narrow: make the current tool truthful, recoverable, and tolerant of common alias variants first.
What changed
src/agents/pi-tools.host-edit.tssrc/agents/pi-tools.read.tssrc/agents/pi-tools.params.tssrc/agents/pi-tools.read.host-edit-recovery.test.tssrc/agents/pi-tools.create-openclaw-coding-tools.adds-claude-style-aliases-schemas-without-dropping-f.test.tsCHANGELOG.mdLinked issues
Test plan
corepack pnpm exec vitest run --maxWorkers=1 src/agents/pi-tools.read.host-edit-recovery.test.tscorepack pnpm exec vitest run --maxWorkers=1 src/agents/pi-tools.create-openclaw-coding-tools.adds-claude-style-aliases-schemas-without-dropping-f.test.tscorepack pnpm exec oxlint src/agents/pi-tools.host-edit.ts src/agents/pi-tools.read.ts src/agents/pi-tools.params.ts src/agents/pi-tools.read.host-edit-recovery.test.ts src/agents/pi-tools.create-openclaw-coding-tools.adds-claude-style-aliases-schemas-without-dropping-f.test.tsNotes
I accidentally invoked the repo's
pnpm testwrapper once while validating this, which routes throughscripts/test-parallel.mjsonmb-air. I did not rely on that run and reran the focused files with direct serial Vitest invocations above.