runAutoRetire

runAutoRetire(deps): RetiredRecord[]

Defined in: src/learnings-lifecycle.ts:149

Cross-repo sweep: identify and retire underperforming active learnings.

For each repo path:

1. Fetch active+promoted rules (injected set) and retired rules.
2. Compute the base rate using all rules (incl. retired) to prevent cascade.
3. Filter to ACTIVE-only candidates (promoted rules keep a verbatim CLAUDE.md copy; retiring them is inert — leave them alone).
4. Sort worst-first (lowest Wilson bound) so the budget cap retires the worst rules.
5. For each shouldRetire candidate:
   • If autoOptimizeFlagged and not yet optimized: optimizeOne (no retire, no budget hit).
   • Else if budget remains: retire it and increment the per-repo counter.

Returns flat array of RetiredRecord across all repos.

Parameters

deps

AutoRetireDeps

Returns

RetiredRecord[]