mandate-2-4-3-code-evaluation

name	mandate-2-4-3-code-evaluation
description	Evaluate compliance for Mandate 2.4.3 (Prevention of Unintended Consequences) using repository code and configuration analysis. Use when producing deterministic pass/fail findings with severity, mapped mitigated vulnerabilities, and file-level evidence.

Mandate 2.4.3 Code Evaluation Skill

Mandate

• ID: 2.4.3
• Title: Prevention of Unintended Consequences

Mitigates

• LLM06 Excessive Agency
• ASI08 Cascading Failures
• ASI10 Rogue Agents

Inputs

• Agent orchestration loops
• KPI/health monitoring hooks
• Circuit breaker modules
• Fallback/kill-switch handlers

Workflow

1. Identify autonomous execution loops.
   • Locate schedulers, planners, retry loops, multi-step agent runners.
2. Verify hard-coded boundaries.
   • Check max iterations, max tool calls, max runtime, max spend/token budget, max side-effect count.
3. Verify system KPI guard checks.
   • Confirm critical KPI thresholds are evaluated before continuing autonomous execution.
4. Verify circuit breaker trigger logic.
   • Confirm threshold breaches trigger immediate halt or degrade-to-safe mode.
5. Verify fail-closed behavior.
   • On missing metrics/errors/timeouts, execution should stop or require human intervention.
6. Verify manual override path.
   • Confirm emergency kill switch can halt active runs.
7. Verify post-trigger handling.
   • Confirm alerting/logging captures trigger reason and execution context.
8. Emit findings per loop with missing boundary or breaker controls.

Decision Rules

• Pass if all autonomous loops have enforceable hard limits and fail-closed circuit breakers tied to system health/KPIs.
• Fail if any autonomous path can run without these controls.

Severity Rules

• Critical: no circuit breaker for high-impact autonomous loop.
• High: boundaries exist but fail-open behavior.
• Medium: controls exist but weak observability.

Output Template

{
  "mandate_id": "2.4.3",
  "status": "fail",
  "severity": "critical",
  "vulnerability_tags": ["LLM06", "ASI08", "ASI10"],
  "evidence": [
    {"file": "src/agent/runner.ts", "line": 102, "detail": "Autonomous loop has no max-iteration boundary"},
    {"file": "src/agent/circuitBreaker.ts", "line": 44, "detail": "Breaker defaults to allow on KPI read failure"}
  ],
  "remediation": "Add hard execution limits and fail-closed breaker logic with mandatory halt on KPI anomalies."
}

Guardrails

• Use code and repo configuration evidence only.
• Always include file-level evidence for each finding.
• If required evidence is missing in code, mark control as not implemented.