Final quality audit
Skill ahtishamshahzad/agent_dev_flow/.ai/skills/final-quality-audit
Use as the last gate before release to confirm the work is genuinely complete, correct, tested, secure, documented, and in scope. Aggregates the specialist reviews into one go/no-go verdict; blocks on any failed gate.From its SKILL.md
npx -y skills add ahtishamshahzad/agent_dev_flow --skill final-quality-auditAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 25 days oldThe repository was created 25 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 0 stars0 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
3.7 KB, 763 tokens by cl100k_base, as published. Nobody here has run it
Final Quality Audit
Purpose
Provide the final go/no-go before release by confirming all gates are genuinely satisfied — correctness, tests, security, docs, scope — and aggregating specialist reviews into one verdict. Supports Gates 6–7.
When to Use
- Immediately before release/merge of a completed work item.
- As the closing check of the orchestration pipeline.
- Not as a replacement for the specialist reviews it aggregates.
Inputs
- The completed change set and its work item/acceptance criteria.
- Outputs of
code-review,security-review,performance-review,testing-strategy,ai-output-review.
Discovery Questions
- Do results meet the original acceptance criteria and success criteria?
- Are Gates 5–6 genuinely passed (not assumed)?
- Is anything out of scope or incomplete?
- Are unrun checks explicitly flagged?
Responsibilities
- Confirm the work meets acceptance/success criteria.
- Confirm tests exist and pass (or are flagged) — Gate 5.
- Confirm security review done, no open Critical/High — Gate 6.
- Confirm docs/changelog updated.
- Confirm scope wasn't expanded and nothing is half-done.
- Aggregate specialist findings into one verdict; block on any failure.
Required Workflow
- Re-read acceptance/success criteria.
- Collect specialist review outputs.
- Verify each gate's evidence (not claims).
- Check scope and completeness.
- Issue a go/no-go verdict with rationale; list blockers if no-go.
Decision Rules
- Any open Critical/High security issue → no-go.
- Missing required tests or failing tests → no-go.
- Unmet acceptance criteria or half-done work → no-go.
- Scope expansion beyond approval → flag and split, don't wave through.
- Evidence over claims: "done" requires verification.
Rules
- Aggregate, don't duplicate the specialists' work — rely on their reports.
- Be honest: distinguish verified from assumed.
- Block on failed gates; never override a gate to ship.
Anti-Patterns
- Rubber-stamping without checking gate evidence.
- Passing work with open Critical/High issues.
- Accepting "should work" as done.
- Letting scope creep through unremarked.
Validation Checklist
- Acceptance/success criteria met.
- Tests present/passing (or flagged) — Gate 5.
- Security review done; no open Critical/High — Gate 6.
- Docs/changelog updated.
- Scope respected; nothing half-done.
- Single go/no-go verdict with rationale.
Definition of Done
A final go/no-go verdict backed by verified gate evidence and aggregated specialist findings; on no-go, a clear, prioritized blocker list.
Related Skills
code-review, security-review, performance-review, testing-strategy, ai-output-review, documentation, release-planning, project-orchestrator.
Related Knowledge
../../knowledge/, ../../projects/current/ (gate status).
Related References
../../checklists/ (gate checklists).
Context Loading Guidance
- Requires: acceptance criteria, gate status, specialist review outputs.
- Does not require: re-reading all source (rely on specialist reports), unrelated references.
- May load: any specialist report needed to confirm a gate.
- Stop when: the go/no-go verdict is issued.
Token Efficiency Guidance
Consume specialist summaries rather than re-deriving them. Verify the decisive evidence per gate; don't re-review the whole change from scratch.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.