AgentStack
Sign in Publish
MCP servers Skills Toolkits Why AgentStack Sell Docs
Sign in
SKILL verified MIT Self-run

Audit Loop

skill-minghinmatthewlam-agent-guards-audit-loop · by minghinmatthewlam

Lightweight read-only audit. Use host-appropriate review coverage for evidence and findings. Iterate on disagreement or low confidence. No code changes.

↗ Repository
— No reviews yet
0 installs
17 views
0.0% view→install

Install

$ agentstack add skill-minghinmatthewlam-agent-guards-audit-loop

✓ scanned · ✓ verified — works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • ✓ Prompt-injection patterns
  • ✓ Secret / credential exfiltration
  • ✓ Dangerous shell & filesystem operations
  • ✓ Untrusted network calls
  • ✓ Known-malicious package signatures

What it can access

  • ✓ Network access No
  • ✓ Filesystem access No
  • ✓ Shell / process execution No
  • ✓ Environment & secrets No
  • ✓ Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

Are you the author of Audit Loop? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

Audit Loop

Read-only audit with review coverage appropriate to the current host. No code changes.

Planning before implementation → use plan-loop instead. Reviewing completed code changes → use review-loop instead.

Prerequisites

Read ~/dev/agent-guards/AGENTS.md before starting — the safety and workflow conventions that govern all agent work.

How It Works

  1. Investigate the question. Collect evidence (files, commands, source material).
  2. Review your findings per references/review-protocol.md. Spin up all review agents in parallel using the host-specific coverage rules. Pre-warm any deferred tools only when the selected host path actually needs them.
  3. Iterate if reviewers disagree, confidence is low, or key claims lack evidence.
  4. Write a final verdict: scope reviewed, evidence basis, findings, residual risks, next steps.

Every meaningful claim must cite specific files, commands, or source material.

Round Sizing

  • Claude host:
  • R1: 1 native + 1 Codex cross-model.
  • R2+ only when triggered: confidence below 85, reviewer disagreement, or high-impact finding lacks evidence. Focus on previous round's findings and unresolved gaps — not a fresh re-review.
  • Escalate to 3+1 for high-risk or persistent disagreement.
  • Codex host:
  • R1: 1-2 native reviewers only.
  • R2+ only when triggered: confidence below 85, reviewer disagreement, or high-impact finding lacks evidence. Focus on previous round's findings and unresolved gaps — not a fresh re-review.
  • Escalate to 3 native reviewers for high-risk or persistent disagreement.

Confidence

After each round, assess confidence 0-100 with top unknowns and what evidence is still missing:

  • Below 85: consider another round.
  • 85+: eligible to close.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet — be the first.

Write a review

Versions

  • v0.1.0 Imported from the upstream source.

More in Agent Skills