- Add currentModel to callback handler for checkmark display
- Add 64-byte callback_data limit protection (skip long model IDs)
- Add tests for large model lists and callback_data limits
* feat: Make BlueBubbles the primary iMessage integration
- Remove old imsg skill (skills/imsg/SKILL.md)
- Create new BlueBubbles skill (skills/bluebubbles/SKILL.md) with message tool examples
- Add keep-alive script documentation for VM/headless setups to docs/channels/bluebubbles.md
- AppleScript that pokes Messages.app every 5 minutes
- LaunchAgent configuration for automatic execution
- Prevents Messages.app from going idle in VM environments
- Update all documentation to prioritize BlueBubbles over legacy imsg:
- Mark imsg channel as legacy throughout docs
- Update README.md channel lists
- Update wizard, hubs, pairing, and index docs
- Update FAQ to recommend BlueBubbles for iMessage
- Update RPC docs to note imsg as legacy pattern
- Update Chinese documentation (zh-CN)
- Replace imsg examples with generic macOS skill examples where appropriate
BlueBubbles is now the recommended first-class iMessage integration,
with the legacy imsg integration marked for potential future removal.
* refactor: Update import paths and improve code formatting
- Adjusted import paths in session-status-tool.ts, whatsapp-heartbeat.ts, and heartbeat-runner.ts for consistency.
- Reformatted code for better readability by aligning and grouping related imports and function parameters.
- Enhanced error messages and conditional checks for clarity in heartbeat-runner.ts.
* skills: restore imsg skill and align bluebubbles skill
* docs: update FAQ for clarity and formatting
- Adjusted the formatting of the FAQ section to ensure consistent bullet point alignment.
- No content changes were made, only formatting improvements for better readability.
* style: oxfmt touched files
* fix: preserve BlueBubbles developer reference (#8415) (thanks @tyler6204)
The previous migration to tsdown was reverted because it caused a ~20x slowdown when running OpenClaw from the repo. @hyf0 investigated and found that simply renaming the `dist` folder also caused the same slowdown. It turns out the Plugin script loader has a bunch of voodoo vibe logic to determine if it should load files from source and compile them, or if it should load them from dist. When building with tsdown, the filesystem layout is different (bundled), and so some files weren't in the right location, and the Plugin script loader decided to compile source files from scratch using Jiti.
The new implementation uses tsdown to embed `NODE_ENV: 'production'`, which we now use to determine if we are running OpenClaw from a "production environmen" (ie. from dist). This removes the slop in favor of a deterministic toggle, and doesn't rely on directory names or similar.
There is some code reaching into `dist` to load specific modules, primarily in the voice-call extension, which I simplified into loading an "officially" exported `extensionAPI.js` file. With tsdown, entry points need to be explicitly configured, so we should be able to avoid sloppy code reaching into internals from now on. This might break some existing users, but if it does, it's because they were using "private" APIs.
grammY getUpdates returns "Request to getUpdates timed out after 500 seconds"
but RECOVERABLE_MESSAGE_SNIPPETS only had "timeout". Since
"timed out".includes("timeout") === false, the error was not classified as
recoverable, causing the polling loop to exit permanently.
Add "timed out" to RECOVERABLE_MESSAGE_SNIPPETS so the polling loop retries
instead of dying silently.
Fixes#7239Fixes#7255
* fix: skip audio files from text extraction early
Audio files should not be processed through extractFileBlocks for text
extraction - they are handled by the dedicated audio transcription
capability (STT).
Previously, audio files were only skipped if they didn't "look like text"
(looksLikeUtf8Text check). This caused issues where some audio binary
data (e.g., long Telegram voice messages) could accidentally pass the
heuristic check and get processed as text content.
This fix:
1. Adds audio to the early skip alongside image/video (more efficient)
2. Removes the redundant secondary check that had the flawed condition
Fixes audio binary being incorrectly processed as text in Telegram and
other platforms.
* Media: skip binary media in file extraction (#7475) (thanks @AlexZhangji)
---------
Co-authored-by: Shakker <shakkerdroid@gmail.com>
Address P1 review feedback from Greptile: instanceof AbortSignal may be
unreliable across different realms (VM, iframe, etc.) where the AbortSignal
constructor may differ. Use structural typing (checking for aborted property
and addEventListener method) for more robust cross-realm compatibility.
* feat(config): add subagent default thinking
* fix: accept config subagents.thinking + stabilize test mocks (#7372) (thanks @tyler6204)
* fix: use findLast instead of clearAllMocks in test (#7372)
* fix: correct test assertions for tool result structure (#7372)
* fix: remove unnecessary type assertion after rebase
* fix(telegram): handle Grammy HttpError network failures (#3815)
Grammy wraps fetch errors in an .error property (not .cause). Added .error
traversal to collectErrorCandidates in network-errors.ts.
Registered scoped unhandled rejection handler in monitorTelegramProvider
to catch network errors that escape the polling loop (e.g., from setMyCommands
during bot setup). Handler is unregistered when the provider stops.
* fix(telegram): address review feedback for Grammy HttpError handling
- Gate .error traversal on HttpError name to avoid widening search graph
- Use runtime logger instead of console.warn for consistency
- Add isGrammyHttpError check to scope unhandled rejection handler
- Consolidate isNetworkRelatedError into isRecoverableTelegramNetworkError
- Add 'timeout' to recoverable message snippets for full coverage
* Fix subagent announce race and timeout handling
Bug 1: Subagent announce fires before model failover retries finish
- Problem: CLI provider emitted lifecycle error on each attempt, causing
subagent registry to prematurely call beginSubagentCleanup() and announce
with incorrect status before failover retries completed
- Fix: Removed lifecycle error emission from CLI provider's attempt-level
.catch() in agent-runner-execution.ts. Errors still propagate to
runWithModelFallback for retry, but no intermediate lifecycle events
are emitted. Only the final outcome (after all retries) emits lifecycle
events.
Bug 2: Hard 600s per-prompt timeout ignores runTimeoutSeconds=0
- Problem: When runTimeoutSeconds=0 (meaning 'no timeout'), the code
returned the default 600s timeout instead of respecting the 0 setting
- Fix: Modified resolveAgentTimeoutMs() to treat 0 as 'no timeout' and
return a very large timeout value (30 days) instead of the default.
This avoids setTimeout issues with Infinity while effectively providing
unlimited time for long-running tasks.
* fix: emit lifecycle:error for CLI failures (#6621) (thanks @tyler6204)
* chore: satisfy format/lint gates (#6621) (thanks @tyler6204)
* fix: restore build after upstream type changes (#6621) (thanks @tyler6204)
* test: fix createSystemPromptOverride tests to match new return type (#6621) (thanks @tyler6204)
* feat: Implement paragraph boundary flushing in block streaming
- Added `flushOnParagraph` option to `BlockReplyChunking` for immediate flushing on paragraph breaks.
- Updated `EmbeddedBlockChunker` to handle paragraph boundaries during chunking.
- Enhanced `createBlockReplyCoalescer` to support flushing on enqueue.
- Added tests to verify behavior of flushing with and without `flushOnEnqueue` set.
- Updated relevant types and interfaces to include `flushOnParagraph` and `flushOnEnqueue` options.
* fix: Improve streaming behavior and enhance block chunking logic
- Resolved issue with stuck typing indicator after streamed BlueBubbles replies.
- Refactored `EmbeddedBlockChunker` to streamline fence-split handling and ensure maxChars fallback for newline chunking.
- Added tests to validate new chunking behavior, including handling of paragraph breaks and fence scenarios.
- Updated changelog to reflect these changes.
* test: Add test for clamping long paragraphs in EmbeddedBlockChunker
- Introduced a new test case to verify that long paragraphs are correctly clamped to maxChars when flushOnParagraph is enabled.
- Updated logic in EmbeddedBlockChunker to handle cases where the next paragraph break exceeds maxChars, ensuring proper chunking behavior.
* refactor: streamline logging and improve error handling in message processing
- Removed verbose logging statements from the `processMessage` function to reduce clutter.
- Enhanced error handling by using `runtime.error` for typing restart failures.
- Updated the `applySystemPromptOverrideToSession` function to accept a string directly instead of a function, simplifying the prompt application process.
- Adjusted the `runEmbeddedAttempt` function to directly use the system prompt override without invoking it as a function.
* Security: cap Slack media downloads and validate Slack file URLs
* Security: relax web media fetch cap for compression
* Fixes: sync pi-coding-agent options
* Fixes: align system prompt override type
* Slack: clarify fetchImpl assumptions
* fix: respect raw media fetch cap (#6639) (thanks @davidiach)
---------
Co-authored-by: Peter Steinberger <steipete@gmail.com>
Add AbortSignal.timeout() to both fetch calls in download.ts to prevent
indefinite hangs when Telegram API is slow or unresponsive.
- getTelegramFile(): 30s timeout for metadata API call
- downloadTelegramFile(): 60s timeout for file download
Both functions now accept optional timeoutMs parameter for configurability.
Fixes#6849
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Update @mariozechner/pi-agent-core, pi-ai, pi-coding-agent, pi-tui to 0.51.0
- Delete src/types/pi-coding-agent.d.ts (declared additionalExtensionPaths which SDK never supported)
- Fix ToolDefinition.execute signature (parameter order changed in 0.51.0)
- Remove dead additionalExtensionPaths from createAgentSession calls
- Added systemPrompt for overriding the default system prompt.
- Introduced skills for pre-loaded skills management.
- Added contextFiles for handling pre-loaded context files with path and content attributes.
* Fix missing before_tool_call hook integration
- Add hook call in handleToolExecutionStart before tool execution begins
- Support parameter modification via hookResult.params
- Support tool call blocking via hookResult.block with custom blockReason
- Fix try/catch logic to properly re-throw blocking errors using __isHookBlocking flag
- Maintain tool event consistency by emitting start/end events when blocked
- Addresses GitHub issue #6535 (1 of 8 unimplemented hooks now working)
Co-Authored-By: Claude Sonnet 4 <noreply@anthropic.com>
* Add comprehensive test suite for before_tool_call hook
- 9 tests covering all hook scenarios: no hooks, parameter passing, modification, blocking, error handling
- Tests tool name normalization and different argument types
- Verifies proper error re-throwing and logging behavior
- Maintained in fork for regression testing
* Fix all issues identified by Greptile code review
Address P0/P1/P3 bugs:
P0 - Fix parameter mutation crash for non-object args:
- Normalize args to objects before passing to hooks (maintains hook contract)
- Handle parameter merging safely for both object and non-object args
P1 - Add missing internal state updates when blocking tools:
- Set toolMetaById metadata like normal flow
- Call onAgentEvent callback to maintain consistency
- Emit events in same order as normal tool execution
P1 - Fix test expectations to match implementation reality:
- Non-object args normalized to {} for hook params (not passed as-is)
- Add test for safe parameter modification with various arg types
- Update mocks to verify state updates when blocking
P3 - Replace magic __isHookBlocking property with dedicated ToolBlockedError class:
- More robust error handling without property collision risk
- Cleaner control flow that's serialization-safe
Co-Authored-By: Claude Sonnet 4 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4 <noreply@anthropic.com>
* security(web): sanitize WhatsApp accountId to prevent path traversal
Apply normalizeAccountId() from routing/session-key to
resolveDefaultAuthDir() so that malicious config values like
"../../../etc" cannot escape the intended auth directory.
Fixes#2692
* fix(web): check sanitized segment instead of full path in Windows test
* style(web): fix oxfmt formatting in accounts test
highlightMatch() was replacing tokens inside ANSI escape codes,
corrupting sequences like [38;2;123;127;135m when searching for '2'.
Fix: apply highlighting to plain text before theme styling.
- Update @mariozechner/pi-ai and pi-agent-core to 0.50.9
- Rename cacheControlTtl to cacheRetention with values none/short/long
- Add backwards compatibility mapping: 5m->short, 1h->long
- Remove dead OpenRouter check (uses openai-completions API)
- Default new configs to cacheRetention: short
* Slash new: use agent personality in session greeting
Previously /new and /reset used a generic greeting prompt. Agents with
personality files (IDENTITY.md, SOUL.md, etc) would respond out of
character until the conversation got going.
Now the prompt instructs the agent to greet users as their character,
using their defined voice, mannerisms, and mood from the start.
* Auto-reply: avoid workspace references in reset prompt
* fix: avoid workspace references in reset greeting (#5706) (thanks @bravostation)
---------
Co-authored-by: MoltBot <bot@moltbot.com>
Co-authored-by: Shadow <shadow@clawd.bot>
On Windows, non-.exe commands like npm, pnpm, yarn, npx require
their .cmd extension when using spawn(). This adds a resolveCommand()
helper that automatically appends .cmd on Windows for these commands.
Fixes#5773
* feat(routing): add thread parent binding inheritance for Discord
When a Discord thread message doesn't match a direct peer binding,
now checks if the parent channel has a binding and uses that agent.
This enables multi-agent setups where threads inherit their parent
channel's agent binding automatically.
Changes:
- Add parentPeer parameter to ResolveAgentRouteInput
- Add binding.peer.parent match type
- Resolve thread parent early in Discord preflight
- Pass parentPeer to resolveAgentRoute for threads
Fixes thread routing in Discord multi-agent configurations where
threads were incorrectly routed to the default agent instead of
inheriting from their parent channel's binding.
* ci: trigger fresh macOS runners
* Discord: inherit thread bindings in reactions
* fix: add changelog for thread parent binding (#3892) (thanks @aerolalit)
---------
Co-authored-by: Lalit Singh <lalit@clawd.bot>
Co-authored-by: OSS Agent <oss-agent@clawdbot.ai>
Co-authored-by: Shadow <shadow@clawd.bot>
* fix(security): restrict inbound media staging to media directory
* docs: update MEDIA path guidance for security restrictions
- Update agent hint to warn against absolute/~ paths
- Update docs example to use https:// instead of /tmp/
---------
Co-authored-by: Evan Otero <evanotero@google.com>
Small model testing showed the label did not meaningfully help:
- Sub-3B models fail regardless of format
- 8B models untested with label specifically
- Frontier models never needed it
The bracket convention [Wed 2026-01-28 22:30 EST] matches existing
channel envelope format and is widely present in training data.
Saves ~2-3 tokens per message vs the labeled version.
Changes [Wed 2026-01-28 20:30 EST] to [Current Date: Wed 2026-01-28 20:30 EST].
Tested with qwen3-1.7B: even with DOW in the timestamp, the model
ignored it and tried to compute the day using Zeller's Congruence.
The "Current Date:" semantic label is widely present in training data
and gives small models the best chance of recognizing the timestamp
as authoritative context rather than metadata to parse.
Cost: ~18 tokens per message. Prevents hallucination spirals that
burn hundreds or thousands of tokens on date derivation.
Changes [2026-01-28 20:30 EST] to [Wed 2026-01-28 20:30 EST].
Costs ~1 extra token but provides day-of-week for smaller models
that can't derive DOW from a date. Frontier models already handle
it, but this is cheap insurance for 7B-class models.
Replace verbose formatUserTime (Wednesday, January 28th, 2026 — 8:30 PM)
with the same formatZonedTimestamp used by channel envelopes (2026-01-28
20:30 EST). This:
- Saves ~4 tokens per message (~7 vs ~11)
- Uses globally unambiguous YYYY-MM-DD 24h format
- Removes 12/24h config option (always 24h, agent-facing)
- Anchors envelope detection to the actual format function — if channels
change their timestamp format, our injection + detection change too
- Adds test that compares injection output to formatZonedTimestamp directly
Exported formatZonedTimestamp from auto-reply/envelope.ts for reuse.
Verifies that America/New_York correctly resolves to midnight for
both EST (winter, UTC-5) and EDT (summer, UTC-4) using the same
IANA timezone. Intl.DateTimeFormat handles the DST transition.
The chat.send handler (used by webchat and TUI) is a separate path
from the agent handler. Inject timestamp into BodyForAgent (what the
model sees) while keeping Body raw for UI display.
This completes timestamp coverage for all non-channel paths:
- agent handler: spawned subagents, sessions_send, heartbeats
- chat.send: webchat, TUI
Messages arriving through the gateway agent method (TUI, web, spawned
subagents, sessions_send, heartbeats) now get a timestamp prefix
automatically. This gives all agent contexts date/time awareness
without modifying the system prompt (which is cached for stability).
Channel messages (Discord, Telegram, etc.) already have timestamps
via envelope formatting in a separate code path and never reach
the agent handler, so there is no double-stamping risk.
Cron jobs also inject their own 'Current time:' prefix and are
detected and skipped.
Extracted as a pure function (injectTimestamp) with 12 unit tests
covering: timezone handling, 12/24h format, midnight boundaries,
envelope detection, cron detection, and empty messages.
Integration test verifies the agent handler wires it in correctly.
Closes#3658
Refs: #1897, #1928, #2108
Ayaan Zaidi
Ermenegildo Fiorito
Tyler Yust
Peter Steinberger
Gustavo Madeira Santana
Armin Ronacher
Josh Palmer
cpojer
Shakker
Vignesh Natarajan
Benjamin Jesuiter
Justin
mac mimi
Ji
bqcfjwhz85-arch
Elarwei
Mariano Belinky
Mariano Belinky
Christian Klotz
David Iach
chenglun.hu
Sk Akram
Tyler Yust
Mario Zechner
Hasan FLeyah
VACInc
Loganaden Velvindron
Ryan Nelson
Leszek Szpunar
Alex Atallah
Ayaan Zaidi
Evan
clawdinator[bot]
Vignesh
xiaose
bravostation
Jhin
Lalit Singh
Tak Hoffman
Shadow
Glucksberg
Conroy Whitney