agentsclimarketplace
Sign in

Audit agent plan

Skill avksp/agent-lifecycle-kit/skills/audit-agent-plan

Independently audit SDD specifications and agent-ready plans before freeze, returning findings, readiness verdicts, and required refinements.From its SKILL.md

Install
npx -y skills add avksp/agent-lifecycle-kit --skill audit-agent-plan

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 18 stars18 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.1 KB, as published. Nobody here has run it

Audit agent plan

Use this skill for independent review of a candidate specification or agent-ready plan. The default action is audit only.

Inputs

Read the developer overview first, then the specification, plan manifest, write-set contract, worker DAG, acceptance checklist, evidence plan, budget profile, and previous review if this is a revision.

Review matrix

Check every item for:

  • traceability from requirement to acceptance criterion, workstream, evidence, and final audit gate;
  • SDD tier correctness;
  • missing clarification or unsafe assumption;
  • architecture, security, performance, data, browser, release, or external environment risks;
  • exact ownership and forbidden writes;
  • dependency DAG correctness and parallel safety;
  • validation commands that are executable, scoped, bounded, and sufficient;
  • context and token budget fit for small and large context hosts;
  • deterministic SDD tier resolution evidence or a justified higher-tier override;
  • plan/freeze/run artifact separation;
  • adapter neutrality and absence of project-specific assumptions unless the target project explicitly owns them.
  • optional Review Mesh scope: any blocking multi-review requirement must be off by default, provider-neutral, token/resource capped, phase-scoped, and backed by assignment/result/synthesis/quorum evidence routes.

When available, run agent-lifecycle plan completeness-check --manifest <plan.manifest.json> and treat agent-plan-completeness-validation.v1 blockers as deterministic pre-audit findings. For freeze gates that opt in to structural enforcement, also check agent-lifecycle plan check --require-completeness. For Markdown-only plan packages or plan folders, use the cookbook path in docs/guides/lifecycle-cookbook.md and docs/guides/code-review-workflows.md to keep the review draft-only until a real ALK plan is frozen.

Verdicts

  • READY_TO_FREEZE: no open Medium or High finding; freeze inputs are complete.
  • CHANGES_REQUIRED: the plan can be corrected within the current intent.
  • BLOCKED: external authority or missing product decision prevents a safe plan.
  • REOPEN_REQUIRED: a frozen plan needs contract, ownership, or intent changes.

Rules

  • Findings first; do not hide blockers in prose.
  • Do not rewrite the plan during independent review unless explicitly asked.
  • Do not approve a plan that lacks a developer overview, write ownership, evidence gates, or final audit route.
  • Do not ask for longer prose when the defect is missing structure. Require the smallest tier-appropriate field or receipt route that closes the blocker.
  • Do not accept self-review as independent review.
  • Do not downgrade required runtime, security, or performance evidence to synthetic-only checks unless the plan explicitly narrows the claim.

Output

Return:

  • findings ordered by severity;
  • coverage matrix;
  • missing or excessive scope;
  • validation of budgets and context fit;
  • freeze verdict;
  • exact edits required before the next review.

Keep looking

Skills are one crate of 627,801. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.