Skip to content

[cache-strategy] Fix cache miss in Documentation Unbloat #57760

Description

@github-actions

Summary

Documentation Unbloat declares cache-memory: true but current evidence indicates the workflow is effectively cold-starting its cache state and has no recorded cache hits in recent retained history.

Evidence

  • Workflow: Documentation Unbloat
  • Recent analyzed run: 33457827219
  • Historical index after update: runs_analyzed_30d=5, last_cache_hit_date=null, volatile_key_without_restore=true
  • Prompt evidence from run 33457827219:
    • aw-prompts/prompt.txt:161find /tmp/gh-aw/cache-memory/ -maxdepth 1 -ls
    • aw-prompts/prompt.txt:162cat /tmp/gh-aw/cache-memory/cleaned-files.txt 2>/dev/null || echo "No previous cleanups found"
  • The cache is read from a flat root-level file (/tmp/gh-aw/cache-memory/cleaned-files.txt), but the analyzer has not observed a successful cache-hit signal for this workflow across recent retained runs.

Why this matters

Without a stable cache reuse path, the workflow can re-scan and reconsider previously processed docs, wasting tokens and reducing round-robin progress.

Recommended fix

  1. Store workflow state under a dedicated namespace such as /tmp/gh-aw/cache-memory/documentation-unbloat/.
  2. On startup, log whether prior state was loaded and how many entries were restored.
  3. Append or merge state incrementally instead of replacing it wholesale.
  4. Emit an explicit cache-hit/cache-miss log line so future audits can distinguish a first run from a broken cache restore.

Source

Generated by 📊 Daily Cache Strategy Analyzer · pi · gpt54 · 77.9 AIC · ⌖ 9.18 AIC · ⊞ 8K ·

  • expires on Sep 8, 2026, 10:52 AM UTC-08:00

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions