Review Prompts

Reusable .md prompt files that instruct an AI agent how to review a specific aspect of code, plans, or other artifacts.

When to use this skill

Use this skill when the user wants to:

• Review a pull request, commit, or set of code changes
• Audit code for security vulnerabilities
• Evaluate an implementation plan or design document
• Check architecture for SOLID violations or coupling issues
• Assess code for performance bottlenecks
• Review test quality and coverage gaps
• Apply language-specific best practices (TypeScript, Python, Java, Scala, JavaScript)
• Review prompts or AI skill definitions for prompt engineering quality

Available Review Types

Pre-flight checks

Run the preflight script before first use:

python3 <skill_dir>/scripts/rp_preflight.py

This checks Python version, build-prompt.py presence, shared template, prompt files, and a runtime test of --list in one pass. Each check prints [PASS], [FAIL], or [WARN]. Exit code 0 = all checks passed, 1 = at least one failed.

Building Prompts

Each review type is stored as a lean .md file (Role + Checklist only). Shared sections (Process, Output Format, Constraints) live in prompts/_shared.md. A build script assembles the complete prompt with every section in the right place.

Do not use the raw .md files in prompts/ directly — they are incomplete (missing Process, Output Format, Constraints). Always use build-prompt.py to produce a complete, ready-to-use prompt.

python3 scripts/build-prompt.py <type>              # prints to stdout
python3 scripts/build-prompt.py <type> -o <file>    # writes to file
python3 scripts/build-prompt.py --list              # lists available types

Usage

Standalone

1. Run: python3 scripts/build-prompt.py security
2. Read the assembled prompt output
3. Apply its checklist to the target code/plan/artifact
4. Output findings in the structured severity format

With a sub-agent

Build the prompt and pass the output to the sub-agent — the host agent never reads the prompt itself:

1. Determine which review type(s) are needed
2. Run: python3 scripts/build-prompt.py <type> -o /tmp/<type>-prompt.md
3. Pass the output file path to the sub-agent along with the review target
4. The sub-agent reads the assembled prompt, applies the checklist, and returns structured findings
5. Synthesize findings from multiple sub-agents if running parallel reviews

Example

Standalone review of a security concern:

1. Run: python3 scripts/build-prompt.py security
2. Read the target files: src/auth/login.ts, src/middleware/session.ts
3. Apply the security checklist to these files
4. Output findings in the structured severity format

Parallel sub-agent review of a PR:

1. Pick review types: code-review (always), security (auth changes detected)
2. Run: python3 scripts/build-prompt.py code-review -o /tmp/cr-prompt.md
3. Run: python3 scripts/build-prompt.py security -o /tmp/sec-prompt.md
4. Launch sub-agent #1 with /tmp/cr-prompt.md + the diff
5. Launch sub-agent #2 with /tmp/sec-prompt.md + the diff
6. Collect both results, de-duplicate, synthesize into unified review

Custom Prompts

To add a custom review type:

1. Create a new .md file in the prompts/ directory with Role and Checklist sections only (Process, Output Format, and Constraints are assembled from prompts/_shared.md by the build script)
2. Add a row to the Available Review Types table above with the type name, file path, and a short focus description — this helps the agent pick the right prompt