Auditor Gate

Auditor Gate

Run final acceptance gating as a separate role after judge evaluation.

Role Boundaries

• Audit process and risk posture.
• Validate traceability and policy signals.
• Produce audit.json only.
• Do not implement fixes and do not re-grade functional correctness from scratch.

Required Inputs

• task_id
• handoff.json
• verdict.json
• output path for audit.json

Optional:

• eval-results.json
• explicit audit commands (policy/security/license/quality checks)

Workflow

1) Run audit evidence collection

Run:

python3 <path-to-skill>/scripts/run_audit.py \
  --repo <repo-root> \
  --handoff <artifact-path>/handoff.json \
  --verdict <artifact-path>/verdict.json \
  --output <artifact-path>/audit-results.json \
  --command "python3 -m pip list --format=freeze"

This collects artifact-integrity checks and command outcomes. It also checks task and requirement traceability between handoff.json and verdict.json.

2) Write gate artifact

Run:

python3 <path-to-skill>/scripts/write_audit.py \
  --task-id <task-id> \
  --audit-results <artifact-path>/audit-results.json \
  --output <artifact-path>/audit.json \
  --gate fail \
  --finding "policy::Dependency policy review missing" \
  --required-action "Add dependency review evidence" \
  --risk-level high

3) Validate gate artifact

Run:

python3 <path-to-skill>/scripts/validate_audit.py \
  --input <artifact-path>/audit.json

Decision Rules

• pass:
  • no critical audit failures
  • traceability and policy checks pass or are explicitly accepted
• fail:
  • policy, traceability, or release constraints are violated
• needs-human:
  • missing evidence or ambiguous policy interpretation blocks deterministic decision

Output Rules

• Always produce audit.json.
• Keep findings evidence-based and implementation-neutral.
• Provide concrete required_actions for fail and needs-human.
• Report requirement-coverage traceability explicitly in audit.json.

Resources

scripts/

• scripts/run_audit.py: collect artifact and command-level audit evidence.
• scripts/write_audit.py: synthesize audit.json gate decision.
• scripts/validate_audit.py: validate audit artifact schema.

references/

• references/artifact-contract.md: canonical audit.json structure.
• references/gate-rubric.md: practical criteria for pass/fail/needs-human.