caching-and-incremental

# Caching & incremental audit (meta-skill) The first audit of a 50K-line codebase is slow. The second one shouldn't be. This skill governs cache reuse. ## Cache layout `~/.cache/auditsentry/<repo-fingerprint>/` ``` findings/ <sha256(file-content)>.json # findings for that exact file content <sha256>.skill-versions # skill names + versions used to produce findings manifest.json # mapping file-path → current sha256 ``` **Repo fingerprint** = sha256 of the absolute repo root path (so different checkouts don't collide). ## Procedure (per file F to audit) ### Step 1 — Compute current hash `current_hash = sha256(read F)` ### Step 2 — Check manifest - Look up `manifest[F]` → `cached_hash`. - If `cached_hash == current_hash`: - Load `findings/<current_hash>.json`. - Verify `skill-versions` matches current skill set. - If both match: **return cached findings**, mark file as cached in output. - Otherwise → audit fresh and write to cache. ### Step 3 — Write fresh result After audit completes: - `write findings/<current_hash>.json` - Update `manifest[F] = current_hash` ### Step 4 — Garbage collect Periodically (e.g. weekly): drop `findings/*.json` not referenced from any current manifest. ## Invalidation triggers Re-audit even if hash matches when: - The list of active skills has changed since the cached result - A new vulnerability skill was added that may apply - User passes `--no-cache` or `--rerun` - `.auditsentry.yml` config has ch...