Output Guardrail Critic

When to Use

Use when you have an existing guardrail configuration or execution results and need an evidence-backed review before deployment. Best for finding coverage gaps, threshold calibration issues, and safety gate problems.

Trigger patterns:

• "Review these guardrails"
• "Are our output filters sufficient?"
• "Check guardrail coverage for this agent"
• "Review the safety gates before we deploy"

What It Does

Reviews guardrail configurations and/or execution results with a 6-phase investigation protocol covering:

• Output quality gates (hallucination, citation, format, completeness)
• Safety gates (secret leakage, PII exposure, harmful content, bias)
• Runtime approval gates (human-in-the-loop coverage for high-consequence/irreversible actions)
• Configuration coverage and threshold calibration
• 3 perspectives: security engineer, end user, compliance officer

Invocation

/output-guardrail-critic

Or via meta-router:

/meta-critic "review guardrails for [agent description]"

Input

Provide one or both of:

1. Guardrail configuration (YAML, JSON, or descriptive)
2. Guardrail execution results (trigger logs, false positive reports)

Also helpful:

• Agent purpose and deployment context
• Compliance requirements
• Known false positive complaints

Output

• Quality gate and safety gate review tables
• Configuration coverage assessment
• Multi-perspective findings (security, user, compliance)
• Severity-rated findings with evidence
• Verdict: REJECT / REVISE / ACCEPT-WITH-RESERVATIONS / ACCEPT

Companion Skills

• input-guardrail-planner — when REVISE/REJECT, route back for guardrail redesign
• security-threat-model-planner — if the review reveals the agent needs a broader threat model
• proposal-critic — for reviewing the overall guardrail strategy at a higher level