audit-impl
Skill1 starsChanged 8 months agoMITView on GitHub
Audit a completed implementation against its originating plan(s). Returns GO (merge approved) or NO GO (generates remediation file for retry). Final gate before merge in any implementation pipeline.
In this repository59 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- arch-lens-c4-containerSkill
- arch-lens-concurrencySkill
- arch-lens-data-lineageSkill
- arch-lens-deploymentSkill
- arch-lens-developmentSkill
- arch-lens-error-resilienceSkill
- arch-lens-module-dependencySkill
- arch-lens-operationalSkill
- arch-lens-process-flowSkill
- arch-lens-repository-accessSkill
- arch-lens-scenariosSkill
- arch-lens-securitySkill
- arch-lens-state-lifecycleSkill
- audit-archSkill
- audit-bugsSkill
- audit-cohesionSkill
- audit-defense-standardsSkill
- audit-frictionSkill
- audit-testsSkill
- design-guardsSkill
- dry-walkthroughSkill
- elaborate-phaseSkill
- exp-lens-benchmark-representativenessSkill
- exp-lens-causal-assumptionsSkill
- exp-lens-comparator-constructionSkill
- exp-lens-error-budgetSkill
- exp-lens-estimand-claritySkill
- exp-lens-exploratory-confirmatorySkill
- exp-lens-fair-comparisonSkill
- exp-lens-governance-riskSkill
- exp-lens-iterative-learningSkill
- exp-lens-measurement-validitySkill
- exp-lens-pipeline-integritySkill
- exp-lens-randomization-blockingSkill
- exp-lens-reproducibility-artifactsSkill
- exp-lens-sensitivity-robustnessSkill
- exp-lens-severity-testingSkill
- exp-lens-unit-interferenceSkill
- exp-lens-validity-threatsSkill
- exp-lens-variance-stabilitySkill
- id-slopSkill
- implement-worktreeSkill
- investigateSkill
- make-arch-diagSkill
- make-experiment-diagSkill
- make-groupsSkill
- make-planSkill
- make-reqSkill
- make-scenariosSkill
- mermaidSkill
- plan-arcSkill
- rectifySkill
- review-approachSkill
- update-architectureSkill
- update-reqsSkill
- update-specsSkill
- verify-diagSkill
- verify-proposed-diagSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
19 rules. About 0 are common, 7 are distinctive, and 12 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
This author's own choices. Worth a look.
12
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 13
After implementation completes and tests pass, as the last step before merging
- Distinctive: line 36
Run tests — this skill audits, it does not fix
- Very specific: line 37
Create files outside temp/audit-impl/
- Distinctive: line 41
Use Explore subagents for all file reads and diff retrieval
- Very specific: line 43
Write Dry-walkthrough verified = TRUE as the absolute first line of any remediation file
- Very specific: line 51
Single
.mdfile: no comma, ends in .md → use it directly - Very specific: line 55
Directory: no comma, does not end in .md or .json → glob for *_plan_*.md files in the directory
- Distinctive: line 60
Verify every plan file exists.
- Very specific: line 66
Run: git diff {base_branch}...{branch_name} from the current working directory
- Very specific: line 71
Run: git diff {implementation_ref}..{base_branch} — two-dot, SHA on the left
- Distinctive: line 75
If git reports "unknown revision or path not in the working tree", abort with:
- Very specific: line 79
Path-existence guard: Before issuing a Read call on a path that is not guaranteed to exist (e.g., plan file arguments, temp/investigate/ …
- Distinctive: line 89
All files the plan said it would create, modify, or delete
- Distinctive: line 90
All tests the plan said it would add or modify
- Very specific: line 109
If ref lookup fails: abort with a clear error — "branch ref '{implementation_ref}' not found — it may have been absorbed by a fast-forward …
- Very specific: line 113
If --is-ancestor exits 0 (branch is already an ancestor of base — fully merged): log a warning, then treat this as GO with note:
- Very specific: line 147
NO GO: if any finding is MISSING or CONFLICT.
- Very specific: line 149
GO (with notes): if only ODD findings exist — unexpected additions that do not break correctness.
- Very specific: line 152
GO if all findings are COVERED.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Implementation Audit Skill
Audit a completed implementation against its plan(s) before merge. Identifies gaps, missed requirements, scope creep, and unexpected changes. Produces a GO or NO GO verdict.
When to Use
- After implementation completes and tests pass, as the last step before merging
- In single-plan pipelines: audit the worktree against the plan
- In multi-group pipelines: audit the feature branch against all group plans via manifest
- Standalone:
/audit-impl {plans_input} {implementation_ref} {base_branch}
Arguments
{plans_input} {implementation_ref} {base_branch}plans_input— one of:- A single plan
.mdfile path - A comma-separated list of
.mdplan file paths (no spaces around commas) - A directory containing
*_plan_*.mdfiles - A
manifest_*.jsonfrom/make-groups
- A single plan
implementation_ref— worktree path (if intact), branch name, or commit SHA containing the implementationbase_branch— branch to diff against (default:main)
Critical Constraints
NEVER:
- Modify source files, plan files, or any other files — read-only audit only
- Run tests — this skill audits, it does not fix
- Create files outside
temp/audit-impl/ - Emit a GO verdict when any
MISSINGorCONFLICTfinding exists
ALWAYS:
- Use Explore subagents for all file reads and diff retrieval
- Resolve all plan files before starting (abort early if any are missing)
- Write
Dry-walkthrough verified = TRUEas the absolute first line of any remediation file
Workflow
Step 0 — Parse Arguments
Resolve plans_input:
- Single
.mdfile: no comma, ends in.md→ use it directly - Comma-separated
.mdpaths: value contains,→ split on,, trim whitespace from each token. Validate that each trimmed token ends in.md; log a warning and skip any token that does not. Use each valid token as a plan file path - Directory: no comma, does not end in
.mdor.json→ glob for*_plan_*.mdfiles in the directory manifest_*.json: no comma, ends in.json→ parse it; extractgroups[*].filepaths, resolved relative to the manifest's parent directory
Verify every plan file exists. If any are missing, abort with a clear error listing them.
Determine the diff source from implementation_ref:
- If
implementation_refis an existing directory path:- Extract the branch name:
git -C {implementation_ref} branch --show-current - Run:
git diff {base_branch}...{branch_name}from the current working directory
- Extract the branch name:
- Otherwise, detect whether
implementation_refis a commit SHA or branch name:- Detect SHA:
echo "$implementation_ref" | grep -qE '^[0-9a-f]{40}$' - If SHA:
- Run:
git diff {implementation_ref}..{base_branch}— two-dot, SHA on the left - This shows all commits added to base_branch since the snapshot.
- Run:
- If branch name:
- Run:
git diff {base_branch}...{implementation_ref}— three-dot
- Run:
- If git reports "unknown revision or path not in the working tree", abort with:
"implementation_ref '{implementation_ref}' is neither an existing worktree directory nor a known git ref."
- Detect SHA:
Path-existence guard: Before issuing a Read call on a path that is not guaranteed to
exist (e.g., plan file arguments, temp/investigate/ reports, external file references), use
Glob or ls to confirm the path exists first. This prevents ENOENT errors that cascade into
sibling parallel-call cancellations.
Step 1 — Load Plans via Parallel Subagents
Launch one Explore subagent per plan file in parallel. Each returns:
- Plan title and stated scope
- All files the plan said it would create, modify, or delete
- All tests the plan said it would add or modify
- Key requirements and constraints listed in the plan
Aggregate into a unified requirements inventory.
Step 2 — Load Implementation Diff
Stale branch guard (branch name refs only — skip for SHA refs):
# Only run for branch name refs (not SHA):
if ! echo "$implementation_ref" | grep -qE '^[0-9a-f]{40}$'; then
# Step 1: ref must exist at all
git rev-parse --verify {implementation_ref} 2>/dev/null
# Step 2: branch must not already be fully merged into base
git merge-base --is-ancestor {implementation_ref} {base_branch}
fi- If ref lookup fails: abort with a clear error —
"branch ref '{implementation_ref}' not found — it may have been absorbed by a fast-forward merge before audit-impl ran." - If
--is-ancestorexits 0 (branch is already an ancestor of base — fully merged): log a warning, then treat this as GO with note:"Branch '{implementation_ref}' is already an ancestor of '{base_branch}' — absorbed by fast-forward merge prior to audit. No delta to evaluate; returning GO."
Launch one Explore subagent to retrieve:
git diff {base_branch}...HEAD --stat— file-level summarygit log {base_branch}..HEAD --oneline— commit historygit diff {base_branch}...HEAD— full diff
Step 3 — Audit via Parallel Subagents
Divide the requirements inventory into up to 3 slices. Launch parallel Explore subagents, each receiving its slice and the full diff. Each subagent checks:
- Coverage — Is every file and function the plan named present in the diff?
- Correctness — Does the implementation match the plan's stated intent? Flag inversions, missing logic, or wrong approaches.
- Scope creep — What is in the diff that no plan called for? Flag unexpected files or additions.
- Test coverage — Were the plan's specified tests added?
- Cross-plan conflicts (multi-plan only) — Do any two plans' changes interfere or contradict?
Each subagent returns structured findings:
COVERED— requirement satisfied in the diffMISSING— required change absent from diffODD— change in diff with no plan backingCONFLICT— two plans' implementations interfere with each other
Step 4 — Verdict
NO GO if any finding is MISSING or CONFLICT.
GO (with notes) if only ODD findings exist — unexpected additions that do not break
correctness.
GO if all findings are COVERED.
Step 5 — Output
If GO or GO with notes
Print:
## Audit Result: GO
### Scope Audited
{list of plan files audited}
### Summary
{2–3 sentences on overall implementation quality}
### Notes
{Minor ODD findings — not blockers. Omit section if none.}
### Verdict
MERGE APPROVEDIf NO GO
Generate temp/audit-impl/remediation_{topic}_{YYYY-MM-DD_HHMMSS}.md:
Dry-walkthrough verified = TRUE
# Remediation Plan: {topic}
## Audit Context
Generated by `/audit-impl` after auditing:
{list of original plan files}
## Findings
{For each MISSING and CONFLICT finding:}
### {Finding type}: {short title}
- **Plan reference:** {plan file + section}
- **Expected:** {what the plan specified}
- **Found:** {what the diff shows, or "not present"}
## Remediation Steps
{For each finding, a concrete fix:}
### Fix: {short title}
- **File:** {path}
- **Change:** {what to add, modify, or remove}
- **Requirement:** {plan requirement this satisfies}
## Verification
After remediation:
- Run the project's test suite
- Re-run `/audit-impl` to confirm GOThen print:
## Audit Result: NO GO
### Scope Audited
{list of plan files audited}
### Findings
{Mirror the findings from the remediation file}
### Remediation File
{absolute path to remediation file}
### Verdict
MERGE BLOCKED — feed remediation file to /implement-worktree or /retry-worktreeOutput Location
temp/audit-impl/
└── remediation_{topic}_{YYYY-MM-DD_HHMMSS}.md (written on NO GO only)Related Skills
/implement-worktree— produces the worktree this skill audits/make-groups— produces the manifest this skill accepts asplans_input/dry-walkthrough— validates plans before implementation; audit-impl validates after
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

