You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Fix agent resolution and improve skill definitions (#449)
## Summary
- Add variadic `backups ...string` parameter to `GetAvailable` and
`GetAvailableWithConfig` so configured backup agents are tried before
the hardcoded codex→claude-code→gemini→... fallback chain
- Update all call sites (handleEnqueue, handleFixJob, CI poller, review,
fix, analyze, refine, batch) to resolve the backup agent from config via
`ResolveBackupAgentForWorkflow` and pass it through
- When the backup agent is selected and no explicit model was requested,
use `ResolveBackupModelForWorkflow` for the model instead of the default
model
- Fix backup agent in refine and review workflows inheriting the primary
agent's model instead of resolving its own backup model
- Extract shared `applyModelForAgent()` helper used by both refine and
review for backup-vs-primary model resolution, replacing inline
duplicated logic
- Add regression tests that call `applyModelForAgent` directly to verify
backup agent model inheritance and empty-model guard
- Add invocation guards to review skills preventing false triggers when
users paste existing review results
- Make fix skill check conversation context before running discovery
commands, skipping fetch when findings are already present; require
matching pasted findings to jobs by commit SHA before closing
- Replace "ask to commit" in fix skill with deference to project
CLAUDE.md conventions
- Add general principle to all skills: treat instructions as guidelines,
use conversation context, defer to project-level config on conflicts
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
0 commit comments