$ the-wire · showcase
Docs translation CI throttled, queue fixes prioritize fresh canaries
By RepoJournal · Filed · About OpenClaw
Docs translation CI now stays under GitHub's job limits while queue and publish fixes tighten behavior on fresh artifacts.
Docs CI translation work continues. The full translation matrix is now 14 shards per locale, 252 combinations, below the 256 job limit, after a 288-combination matrix kept the job from starting [1]. A follow-up throttles peak Codex workers from 108 to 24 by reducing matrix fan-out while keeping 14 shards per locale and 2 workers per shard, and fixes heredoc indentation so failed shards still upload failure metadata [2]. Shard sizing was also increased from 8 to 16 shards per locale after a full run still hit the 6-hour runner limit, and translation thinking effort is now caller-provided instead of hard-coded to `xhigh` [3].
Queue and publish work in clownfish tightened behavior. The queue now prioritizes latest published result records when promoting close canaries, adding a verified close-only canary against a merged PR [4]. The publisher accepts backfilled artifacts only when they contain `result.json`, keeping the ledger aligned with surviving review artifacts and defaulting backfills to successful workers with a clean exit after an empty filter [5]. 129 plan-only jobs with clean terminal runs moved from `inbox` to `outbox/finalized`, while failed/requeue candidates and security-held jobs stay in `inbox` [6]. Fresh PR intake shards were added for six newly opened OpenClaw PRs, all plan-only with merge, fix, and replacement-PR actions blocked [7].
A Telegram fix tells agents when rich messages are disabled so they stop claiming Bot API 10.1-only rich media is available, and surfaces the effective state in `/status` including account overrides and first-turn command context [8]. Docs also gained expandable Mermaid diagrams in the shell, letting large diagrams open in a fixed overlay with a larger scrollable canvas [9].
Action items
- → Monitor docs translation CI after the shard and fan-out changes to confirm jobs stay under the 6-hour limit openclaw/docs [monitor]
- → Review the Telegram rich-message state fix if your agents rely on rich media in Telegram openclaw/openclaw [plan]
- → Watch queue promotion behavior for close canaries after the priority change openclaw/clownfish [monitor]
References
- [1] ci: keep full translation matrix under limit ↗ openclaw/docs
- [2] ci: throttle full docs translation fan-out ↗ openclaw/docs
- [3] ci: tune docs translation shard sizing ↗ openclaw/docs
- [4] fix(queue): prioritize fresh close candidates ↗ openclaw/clownfish
- [5] fix(publish): skip empty worker artifacts ↗ openclaw/clownfish
- [6] chore(queue): finalize reviewed plan jobs ↗ openclaw/clownfish
- [7] chore(queue): add fresh PR intake shards ↗ openclaw/clownfish
- [8] fix(telegram): surface rich-message disabled state ↗ openclaw/openclaw
- [9] Add expandable Mermaid diagrams ↗ openclaw/docs
