| name | final-review-and-create-pr |
| description | Run a cross-cutting audit across all code produced for a feature. Use when a feature is complete, all PRs are merged, and the user wants a final review for security issues, logic errors, consistency, and best practices. |
Final Audit
Review all code produced for a feature as a whole, looking for issues that are only visible across the full implementation — not within a single PR.
This audit is not a substitute for per-PR code review. It looks for different things: patterns that were introduced early and replicated incorrectly, inconsistencies between modules, and systemic issues that no single PR reviewer could have caught.
Process
1. Locate the feature
Ask the user for the parent PRD issue. Read it in full, including all child issues and their acceptance criteria.
2. Identify the scope
From the PRD and its issues, build a list of every file that was created or modified as part of this feature. Ask the user to confirm the list is complete and nothing is missing.
3. Explore the implementation
Read all files in scope. For each module, understand:
- What it is responsible for
- What its public interface is
- How it interacts with other modules in scope
- How it interacts with the rest of the codebase outside the feature
Do not audit yet. Build a complete mental model first.
4. Audit for systemic issues
Check for the following categories of issues across the entire implementation. For each finding, note the category, the location, a description of the problem, and a suggested fix.
Consistency
- Do all modules follow the same naming conventions as each other and as the surrounding codebase?
- Are error types, error messages, and error handling patterns consistent across modules?
- Are similar operations implemented in the same way everywhere, or has the same problem been solved differently in different places?
- Are internal API contracts consistent — do callers and implementations agree on types, nullability, and failure modes?
Security
- Is user input validated and sanitised at every entry point?
- Are authentication and authorisation checks applied consistently — not just at the API boundary but at the service layer if appropriate?
- Is sensitive data (tokens, passwords, PII) handled consistently and never logged or exposed?
- Are there injection risks (SQL, command, template) at any layer?
- Are dependencies introduced by this feature up to date and free of known vulnerabilities?
Logic
- Are there race conditions or ordering assumptions that could fail under concurrent use?
- Are all failure modes handled — network errors, partial failures, unexpected null values, empty collections?
- Are there off-by-one errors, incorrect boundary conditions, or incorrect assumptions about data ranges?
- Does the implementation match the acceptance criteria in every child issue, including edge cases?
Best practices
- Is there duplicated logic that should be extracted into a shared module?
- Are there deep modules that were implemented as shallow ones — complex interfaces hiding little functionality?
- Are there any modules that have taken on too many responsibilities and should be split?
- Are tests testing external behaviour or implementation details? Are they brittle?
- Is there dead code introduced by this feature?
5. Prioritise findings
Group findings by severity:
- Critical: security vulnerability, data loss risk, or logic error that will cause incorrect behaviour in production
- High: inconsistency or logic error that will cause problems under realistic conditions
- Medium: best practice violation or inconsistency that will cause maintenance problems
- Low: minor inconsistency or style issue with no functional impact
6. Present the report
Present findings grouped by severity. For each finding:
- Location: file and line range
- Category: which audit category it falls under
- Problem: what is wrong and why it matters
- Suggestion: what to do about it
After presenting all findings, give an overall assessment: is this feature safe to leave in production as-is, or are there critical or high findings that should be resolved before considering the feature done?
Ask the user which findings they want to act on and in what order. Do not make any changes without explicit approval.
7. Save the report
Save the audit report to a file in the format and location specified by the user.
# Audit Report:
Parent PRD: #
Date:
Files in scope:
Summary
<2-3 sentence overall assessment. Is this safe to leave in production? What is the most important thing to fix?>
Critical findings
.
Location: <file>:<line-range>
Category: <Security / Logic / Consistency / Best practices>
Problem:
Suggestion:
High findings
(same structure)
Medium findings
(same structure)
Low findings
(same structure)
No findings
