$ the-wire · showcase
Cold-start Gateway diagnostics stop crying wolf, doctor repairs plugin drift
By RepoJournal · Filed · About OpenClaw · Composed from the cited sources · methodology
3 people shipped this
OpenClaw's day was mostly about not lying to the operator: startup diagnostics that mistook slow cold starts for dead Gateways, a doctor that printed fix instructions instead of fixing, and a shim that overwrote real rate-limit errors with invented retry advice.
fix: avoid false Gateway failures during cold starts openclaw/openclaw
Doctor, deep status, and channel status runs during a slow start no longer report Gateway timeout or unreachable; local diagnostics now share the 60-second readiness allowance, report the observed startup phase, and leave a Gateway that is still starting alone. Explicit timeouts stay bounded and provider usage gets the remaining shared budget.
fix(doctor): repair official plugin drift after manual core installs openclaw/openclaw
After a manual npm core upgrade, official catalog plugins used to sit on the old release while `doctor --fix` only printed update commands. It now brings active catalog-official npm plugins up to the installed OpenClaw release, with recorded non-default tags and pins newer than the plugin cohort still authoritative; unavailable packages warn and third-party plugins are left alone.
fix: reuse cached GitHub responses and recover rate-limited login openclaw/octopool
Login and membership refresh now recover from confirmed primary GraphQL exhaustion, so an exhausted GraphQL quota no longer blocks login when REST can still verify organization membership; permission failures and secondary-limit backoff are preserved. Warm pooled responses also reuse eligible fresh cache entries sooner instead of firing redundant GitHub requests.
fix: honor fresh reads for cached Actions logs openclaw/octopool
A completed Actions log could be served from cached bytes without the log endpoint ever being checked, so a deleted log still looked available during a read the caller had explicitly forced live with `cache-control: max-age=0`. Forced-fresh reads now verify upstream existence, while valid cached bytes stay reusable without another download.
perf(ui): reuse session snapshots across subagent trees openclaw/openclaw
The Control UI reloaded its whole session window whenever active sessions belonged to subagent trees; deployed Gateway evidence showed roughly 6 `sessions.list` requests per second across 49 viewers, ~37 ms median list time over 2,300 rows. Existing child sessions and their ancestors now update immediately without reloading the full roster.
fix(providers): complete failed-acquisition SSH cleanup openclaw/crabbox
The long tail was housekeeping: four octopool cache and error-reporting fixes land alongside crabbox's SSH cleanup context widened to three minutes for GCP deletion, an internal Pond process refactor, and a test fix that waits on a test-owned connection instead of a 20-second expiry.
Action items
