← ClaudeAtlas

guardrailslisted

Evaluate a proposed action against config/guardrails.yaml rules. Returns verdict (allow/warn/deny) with the specific rule and reason.
yolo-labz/linkedin-chrome-copilot · ★ 0 · AI & Automation · score 72
Install: claude install-skill yolo-labz/linkedin-chrome-copilot
# guardrails Advisory checker: given a proposed action description plus relevant metadata, evaluates every enabled rule in `config/guardrails.yaml` and returns a verdict. ## Rules evaluated | Rule | Trigger | |-----------------------|------------------------------------------------------| | no_morning_meetings | action=book_slot + slot_hhmm inside start..end | | hell_yes_threshold | action=progress_pipeline + fit_score < threshold | | async_over_live | action=book_slot when channel in prefer_channels | | send_verify_required | action=send + no verify marker | | no_autosend | action=send + no explicit `--approved` flag | ## Contract ```json { "schema": "guardrails-verdict", "action": "book_slot", "verdict": "deny", "triggered": [ {"rule": "no_morning_meetings", "reason": "..."} ] } ``` ## Invariants - Read-only: never mutates state or files. - Default to **deny** on conflicting advice (operator overrides explicitly). - No Chrome, no network.