mistakes — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited mistakes (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Surface bi-temporal mistake memory from engram's graph. Each mistake has a valid_until timestamp and an invalidated_by_commit reference, so an old mistake auto-expires when its underlying cause is fixed.
Run:
engram mistakes -p $CLAUDE_PROJECT_DIRThe output lists active (still-valid) mistakes with confidence scores. When a mistake matches a query, engram surfaces it with ⚠️ at the top of the context packet, weighted 2.5× to ensure it gets attention.
Use this skill:
To record a new mistake explicitly:
engram learn "Don't pass options.scope=user; SETTINGS_LOCAL takes precedence." -p $CLAUDE_PROJECT_DIRUser: "I'm about to refactor the rate limiter — anything I should know?" You: Run engram mistakes -p $CLAUDE_PROJECT_DIR. Filter mentally for entries related to "rate", "limit", "mutex", or the file paths the user mentions. Surface the matches with their confidence scores: "engram has 3 active mistakes near this code, and one of them flagged this exact module:"
User: "Why does this test keep flaking?" You: Run engram mistakes -p $CLAUDE_PROJECT_DIR. Look for entries about flakiness, timing, race conditions, or the test file's name. If a related mistake exists, quote it verbatim — chances are someone already debugged this.
User: "We just fixed a race condition in the queue worker. Remember it for next time." You: Run engram learn "Queue worker had a race condition: don't read state.queue inside the await; snapshot it first" -p $CLAUDE_PROJECT_DIR. Confirm to the user that the mistake is recorded and will surface with ⚠️ on future Edits to that file.
User: "Has anyone fixed this kind of bug before?" You: Run engram mistakes -p $CLAUDE_PROJECT_DIR. The graph carries 2.5x relevance boost on matching results, so if a similar fix exists, it surfaces near the top.
User: "Clear the mistake list — we've outgrown those warnings" You: Don't. Mistakes are bi-temporal — they auto-expire when their underlying cause is fixed (invalidated_by_commit set automatically). Manual clearing should be the last resort. If a specific mistake is no longer relevant but hasn't auto-expired, the user can edit .engram/graph.db with sqlite3, but recommend living with one extra warning instead.
engram init . first.engram query for those.~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.