safety-checks
bởi leegitw
AI & LLMsTóm tắt
Verify before you trust — model pinning, fallbacks, and runtime safety validation.
Mô tả
Kỹ năng này làm gì
Runtime safety verification suite that consolidates model version pinning, fallback chain validation, cache staleness detection, and cross-session state checks. All operations are local file and metadata reads with no external API calls or data transmission. Consolidates four separate skills into a single aliased command.
Replaces four separate skill invocations with a single aliased suite that integrates directly with HEARTBEAT for automated periodic verification.
Khi nào nên dùng
- Catching model version drift before a scheduled agent run
- Verifying fallback config entries exist after modifying agent settings
- Clearing stale cache entries during periodic workspace maintenance
- Detecting orphaned lock files left by a previously crashed session
- Running automated pre-flight safety checks on every session start via HEARTBEAT
Quy trình ví dụ
User asks: run a full safety check before starting the agent workflow
- 1
Run /sc model to compare active model version string against the pinned value in config
- 2
Run /sc fallback to confirm fallback chain entries are declared in safety-checks.yaml
- 3
Run /sc cache to identify cache entries older than the configured TTL
- 4
Run /sc session to check for stale lock files or orphaned temp artifacts in output/safety/
Four-check report with pass/warn/fail status per check; results written to output/safety/ logs
Cài đặt
clawhub install leegitw/safety-checksChạy lệnh CLI, hoặc dán prompt vào Claude Code, Cursor, Codex hay coding agent bạn đang dùng.