$ the-wire · showcase
Codex history replay fixed, Moonshot China auth loops stopped
By RepoJournal · Filed · About OpenClaw · Composed from the cited sources · methodology
openclaw/openclaw fixed two user-facing bugs: Codex conversations no longer resend already delivered history after a final Telegram message, and selecting Moonshot's China API-key option no longer triggers an install loop.
Codex conversations that ended with a final Telegram message used to replay already delivered history into the native thread when the user continued. OpenClaw now advances history coverage from the successful final `message(action=send)` turn outcome instead, while native hook retention still follows native completion [1].
Moonshot's China API-key option in configure used to reinstall an already-enabled provider. The manifest advertises `moonshot-api-key-cn`, but the published 2026.9.3 runtime's secondary auth method omits that wizard choice ID, so the wizard asks again. Inventory and npm install-record discovery are correct; the fix targets the runtime check [2].
Model operations that already hold a prepared runtime used to lose captured model limits and credentials when they skipped agent discovery. The fix uses the supplied runtime to fill missing model and auth stores before creating empty ones, with explicit caller stores still taking precedence [3]. Release validation also corrected an unpublished candidate being selected as its own upgrade baseline for `openclaw@2026.9.3` [4].
A documentation STE pass normalized the component name to "Gateway" in prose and split run-on sentences joined by semicolons [5]. Five automated doc mirrors from openclaw/openclaw landed in openclaw/docs with no prose changes [6][7][8][9][10].
Action items
- → Update openclaw to pick up the Codex history replay and Moonshot China auth fixes openclaw/openclaw [plan]
- → Re-run model operations that skip agent discovery and rely on captured limits or credentials openclaw/openclaw [monitor]
- → Verify release validation no longer selects an unpublished candidate as its own upgrade baseline openclaw/openclaw [monitor]
References
- [1] fix(codex): avoid replaying history after terminal message delivery ↗ openclaw/openclaw
- [2] fix(configure): stop Moonshot China auth reinstall loops ↗ openclaw/openclaw
- [3] fix(models): reuse prepared stores when discovery is skipped ↗ openclaw/openclaw
- [4] fix(release): omit unpublished candidate from upgrade baselines ↗ openclaw/openclaw
- [5] docs: STE pass on terminology consistency and run-on sentences (#143770) ↗ openclaw/openclaw
- [6] chore(sync): mirror docs from openclaw/openclaw@9d1740187791a9550b838239f9e8fd6dd03ade1c ↗ openclaw/docs
- [7] chore(sync): mirror docs from openclaw/openclaw@d146f3b7d330cab7280131bfceaf2448a2de275f ↗ openclaw/docs
- [8] chore(sync): mirror docs from openclaw/openclaw@b280237832a056d42f88ddd13a75526053da4a1b ↗ openclaw/docs
- [9] chore(sync): mirror docs from openclaw/openclaw@c2af16fed84e1ce9535872bdf3766609e41a3d22 ↗ openclaw/docs
- [10] chore(sync): mirror docs from openclaw/openclaw@43e9b70c474499b837ecde9e3ba2a945261034a5 ↗ openclaw/docs