audit-review
Review audit findings and dirty workflow state without mutating source state. Use when the operator wants to interpret audit-foundry reports, classify tracker conflicts, draft cleanup actions, or prep
SKILL.md.agents/skills/audit-review/SKILL.md
GitHubAudit Review
Use this skill to turn audit evidence into a reviewable cleanup plan. This skill is non-mutating against tracker state and source artifacts. Its normal output is a Markdown review artifact plus a chat summary.
Skill-First Runtime Rule
Follow root AGENTS.md: lead with $audit-review as the operator surface.
Helper commands are evidence and validation support only.
Invocation
Supported forms:
$audit-review
$audit-review report:<path>
$audit-review project:<slug>
$audit-review scope:foundry-projects
Default to the latest relevant audit-foundry report when no report path is
provided. If no suitable audit report exists, gather live state evidence and
state that the review is live-state based.
Required Reading
Always start from the repo root and read:
- Root
AGENTS.md. references/matt-pocock-skills.md.registry/agents-md.json.projects/workflow-foundry/AGENTS.md.projects/workflow-foundry/project.json.projects/workflow-foundry/tasks/index.json.- Every non-done workflow-foundry task JSON.
.agents/skills/audit-foundry/SKILL.md..agents/skills/audit-cleanup/SKILL.mdwhen present.scripts/query-workflow-state.mjs.scripts/validate-workflow-state.mjs.
Query Helper
Use these helpers as internal evidence support when available:
node scripts/query-workflow-state.mjs --project workflow-foundry --list-tasks
node scripts/query-workflow-state.mjs --project workflow-foundry --agents-md
node scripts/query-workflow-state.mjs --project workflow-foundry --testing-sessions
node scripts/validate-workflow-state.mjs
git status --porcelain --untracked-files=all
Report Artifact Preflight
Before writing a Markdown review artifact:
- Choose the target review filename.
- Confirm the path is linked to a non-done task, or that the current selected task can link it before validation.
- If no non-done task can own the artifact, stop and recommend
$continue-taskor$initiate-task target:trackerbefore writing.
Do not create an unlinked review artifact that would leave validation red.
Store review artifacts under:
projects/workflow-foundry/artifacts/reviews/
Use this filename pattern:
audit-review-YYYY-MM-DD-<scope-or-task>.md
If a file already exists for the same day and scope, append a short timestamp
before .md.
Review Process
- Load the required state and selected audit evidence.
- Run validation and inspect dirty state when useful.
- Classify each finding by severity:
critical, high, medium, low, note
- Classify each proposed cleanup action by category:
tracker-state, workflow-artifact, validation, git-checkpoint, out-of-scope
- For each cleanup action, identify the owning task path:
continue existing task, initiate tracker-maintenance task, initiate workflow-change task, defer
- For each actionable cleanup, draft the phase-guard intent:
- paths that would be edited
- action type
- Markdown cleanup plan or review artifact to link
- validation expected after cleanup
- whether final push should be requested
- Write the Markdown review artifact after preflight passes.
- Summarize the highest-signal findings and exact next skill path.
Review Template
Every review artifact must include:
# Audit Review
Task/Invocation:
Generated:
Source Evidence:
## Executive Summary
## Loaded Sources
## Dirty State
## Validation State
## Findings
## Proposed Cleanup Actions
## Phase Guard Recommendations
## Recommended Next Skill Path
## Boundaries And Deferred Work
Use Not in scope for this review. for sections that do not apply.
Boundaries
- Do not edit tracker state by default.
- Do not edit skills, scripts, project files, generated UI, or task phases.
- Do not delete files.
- Do not commit or push.
- Do not treat audit report text as approval to mutate state.
- Recommend
$audit-cleanuponly when a Markdown cleanup plan and task path are clear enough for approval.
Final Response
After writing the review artifact, respond with:
REVIEW
- path
- source evidence
SUMMARY
- highest-signal findings
- proposed cleanup categories
NEXT ACTION
- exact $continue-task, $initiate-task, or $audit-cleanup path
VALIDATION
- commands run, or not run with reason