- fix(gateway): keep conversation delivery within agent bindings — Additional instructions MUST: Keep Allow edits from maintainers enabled for this PR so maintainers can help update the branch when needed. What Problem This Solves Fixes an issue where multi-agent...
- [Bug]: gateway event-loop stalls - 76% of a 7.6s stall is log redaction (replacePatternBounded), one full regex pass per pattern — What happened On a Windows host running six agents, the gateway event loop stalls for 60-200s several times an hour. TCP still accepts on the gateway port so Test-NetConnection reports healthy, but...
- fix(gateway): prevent silent delivery loss on closing connections — What Problem This Solves Fixes an issue where operators using the Gateway or Control UI would silently lose responses, live updates, and worker connection handshakes when a WebSocket entered its...
- fix(codex): agents ignore AGENTS.md on sandboxed app-server runs [AI] — Closes #120600 What Problem This Solves Fixes an issue where agents running through the Codex app-server with a sandbox enabled never receive their AGENTS.md. The workspace file is read and reported...
- fix(together): add size range validation for video dimensions — Problem The Together video generation provider parsed width/height from the size string but did not validate the range. Out-of-range values (e.g. 9999x9999) would be sent to the API, causing...
- fix(cli): return JSON for channel capability failures — What Problem This Solves Fixes an issue where openclaw channels capabilities --json returned exit code 1 with empty stdout for semantic validation failures: --account or --target without a specific...
- refactor(context-engine): remove owner cleanup helper — What Problem This Solves Context-engine registry production code exported an owner-specific cleanup helper solely for test fixture cleanup. Gateway health tests used that helper to delete selected...
- fix(settings): take the settings file lock before the first-write read — Closes #124394 What Problem This Solves Fixes an issue where users running two OpenClaw processes that each save a setting before settings.json exists would silently lose one of those settings. This...
- Dashboard: Anthropic provider group missing from full model settings panel (stale after gateway restart, no config change) — Summary The Anthropic provider group intermittently disappears from the full model settings panel in the OpenClaw Control UI dashboard ("Chat model" settings → provider list), even though: - The...
- Visible sessions render incompletely—UI omits assistant messages despite transcript presence — - Sessions appear in the sidebar as expected. - When opened, parts of the conversation, especially assistant messages, are missing despite being present in the transcript and visible in Mission...
- [Bug]: Codex sessions.compact acquires a second app-server and hits active-writer conflict — Bug type Behavior bug (incorrect output/state without crash) Beta release blocker No Summary Manual sessions.compact on a bound Codex app-server session can acquire a second app-server client for a...
- Feishu: duplicate final replies can occur after model failover from rate-limited primary — Summary In Feishu direct messages, a single user message can sometimes result in two final assistant replies when the primary model fails with rate limiting and OpenClaw falls back to a secondary...
- [Bug]: Main-session compaction holds the session write-lock past the 60s acquire timeout; completed subagent announces time out and give up, silently losing results — Bug type Behavior bug (incorrect output/state without crash) Beta release blocker No Summary When the main session triggers transcript compaction (trigger=cli_budget), the compaction flow holds the...
- PDF fallback applies extraction budgets per file without an aggregate request bound — Summary The PDF tool documents and admits up to ten PDFs. On the generic extraction fallback, each PDF independently receives the full 200,000-character and 4,000,000-render-pixel budgets, then every...
- fix(text): only insert separator between word chars when stripping model tokens — What Problem This Solves Closes #121426. stripModelSpecialTokens inserts a separator space whenever both characters surrounding a removed leaked control token are non-whitespace. It does not...
- A failed tool call always terminates the visible turn as "temporary model failure" — the error is never returned to the model (0/40 recoverable in 24h) — Summary When a tool call fails inside an embedded run, the visible turn terminates immediately. The failure is not returned to the model as a tool result, so the model never gets to adapt, work...
- apply_patch commits earlier hunks before returning a later-hunk error — What happens A valid multi-file apply_patch envelope mutates each hunk immediately. If a later hunk is invalid, mismatched, or cancellation arrives between hunks, the tool rejects with only the later...
- 5.26 supervisor mode + dual-plist install: 30s EADDRINUSE loop + silent-fail unset order — OpenClaw 5.26 supervisor mode + dual-plist install: 30s loop + silent-fail unset order TL;DR: When ai.openclaw.gateway and ai.openclaw.node plists both run, OpenClaw 5.26's detectRespawnSupervisor()...
- [Bug]: sessions.reset RPC on agent:main:main throws "RangeError: Maximum call stack size exceeded" (UNAVAILABLE) in 2026.7.2-beta.7 — Bug type Behavior bug + crash (uncaught JavaScript stack overflow, reproducible via official RPC) Beta release blocker Yes — the official sessions.reset RPC cannot be used against the protected main...
- Record what agents re-read after auto-compaction — What Problem This Solves After auto-compaction summarizes context away, the model often immediately re-reads what was just summarized — re-issuing read/exec calls whose signatures existed right...
- Cron events refresh the table but leave selected job detail stale — What happens When another authorized client updates or removes a selected Cron job, the Control UI receives an exact lifecycle event but ignores its action, jobId, and job snapshot. It refreshes only...
- Feature: Graceful Gateway Restart with Session Recovery — Problem When the gateway restarts — whether from openclaw gateway restart, a config change, SIGUSR1, or a crash — all in-flight work is silently killed. There is no mechanism for sessions to know...
- fix(gateway): keep rootless sessions usable after permission changes — Fixes #128039 What Problem This Solves Fixes an issue where selecting a root-dependent permission mode on a legacy session with no recorded session root could persist an invalid state. Later turns...
- feat(security): require acknowledgement for install policy warnings — New behavior An external security.installPolicy command can now return warn, letting an authorized operator review a suspicious plugin or skill install before deciding whether to continue. -...
- [Bug] Flaky session token validation on clock skew across clustered nodes — Description When validating session tokens generated by peer nodes, small clock skews (>1 second) cause tokens created in the near future to be rejected immediately as invalid. Affected Files -...
- OpenClaw repo stats for 2026-08-23 — 387190 stars, 81313 forks, 3122 contributors
- OpenClaw sponsors (94 total) — 94 sponsors supporting OpenClaw
- fix(publish): ignore duplicate leased recovery dispatches (#3510)
- fix(workers): terminalize skill evaluation outcomes (#3509)
- Add skill: riffkit/riffkit
- OpenClaw — Deploy your own personal AI assistant with OpenClaw on a DigitalOcean Droplet® server, a powerful open-source platform that runs entirely on your infrastructure. OpenClaw connects to the messaging...
- Ask HN: Who is using OpenClaw?
- You are not supposed to install OpenClaw on your personal computer
- Saving 50-70% on Coding Agent Bills: Protecting Prompt Cache Affinity with VMR — TL;DR — With per-token pricing now the default everywhere, the biggest line item in a coding-agent...
- My 262k-token model kept faceplanting at 33k and the fix was nowhere near OpenClaw — If OpenClaw says 262k tokens but ollama ps shows ~33k, the real bottleneck is usually Ollama runtime allocation or Docker env handling, not the mode
- Building Reusable Agent Skills for OpenClaw
- ClawHub Skill Spotlight: AI Workflow Automation Expert — A structured approach to designing and running multi-step AI agent workflows — from simple cron tasks to full multi-agent coordination systems.
- Grok Bot: architecture and design — Grok Bot is early beta. Surfaces and limits can change.Continue reading on Medium »
- Build a Local Code Review & Refactoring Assistant with OpenClaw & Ollama — How to build a private, zero-cost AI code auditor on your own machine in about 30 minutes?Continue reading on ILLUMINATION »