$ the-wire · showcase
ChatGPT-login model picker re-pinned to bundled Codex
By RepoJournal · Filed · About OpenClaw · Composed from the cited sources · methodology
OpenClaw restores the bundled-Codex pin so ChatGPT-login model discovery stops offering models the default runtime rejects, while the rest of the day is internal plumbing: worktree auto-forward, Gateway startup and restart settling, and CI that kills the process group it timed out.
fix(openai): ChatGPT-login picker lists models the bundled Codex rejects openclaw/openclaw
Model discovery for ChatGPT logins reports the bundled Codex version (0.160.0) again, so the picker lists only models the bundled build can run. Previously the picker could offer a model that then failed with HTTP 400 on the default runtime whenever npm's newest Codex ran ahead of OpenClaw's pin; new ChatGPT models show up when the pin moves, as before.
perf(worktrees): unblock default auto-forward during unrelated rebases openclaw/openclaw
Local default-branch auto-forward was being permanently blocked by unrelated paused rebases in linked worktrees; creating a managed worktree now advances an eligible clean local default again. Operations that genuinely reserve the base, explicit bases, remote-base selection, dirty-checkout protection and background preparation behave as before, with no migration or config change.
fix(gateway): settle restart intents and update reports in workers openclaw/openclaw
Restart-intent consumption could block the Gateway signal thread behind a SQLite writer and then hand back an intent whose deletion failed, letting a later signal consume it a second time; consumption is now atomic and runs on the existing workers, as do handoff writes, report receipt transitions and cleanup. Update-report receipts also move off main-thread SQLite so accepted work settles befor...
fix(acp): allow Gateway startup while agent stores are pending openclaw/openclaw
Gateway startup no longer fails before its listener binds when historical ACP metadata points at an agent database still awaiting startup inspection and preparation. Those isolated stores stay unavailable until their existing checks finish, and live legacy ACP bindings still require offline Doctor repair; no schema or config migration is involved.
ci(tests): kill a timed-out test file's whole process group openclaw/openclaw-enterprise
runNode in scripts/ci/run-tests.mjs signaled only the node --test runner on timeout, so the isolated test-file child and anything it started could be re-parented to init still holding a database connection, port or container while lane cleanup ran, which the project measured at roughly once in ten timeouts locally. The runner is now killed by its whole process group.
refactor(review): retire dead decision fields (#1758) openclaw/clawsweeper
The review schema drops liveProofPlan and mantisRecommendation across the prompt, decision type, parser, report document, comment rendering and default decisions. The mantisRecommendation suggestion never triggered anything, since every OpenClaw Mantis workflow is workflow_dispatch-only, and stored reports keep parsing.
