review-commands

Review Commands

This plugin wraps the local CLI with plugins/code-review-assistant/scripts/review.py and uses internal tool skills when needed.

Use this skill when the user asks to:

• run /code-review-assistant:code_analysis <path>
• run /code-review-assistant:add <tool> <report-path>
• add already generated static-analysis reports to a review config
• run /code_analysis collect
• run /code_analysis first-pass
• run /code_analysis source-pass
• run /code_analysis blind-review
• run /code_analysis merge
• run /code_analysis export
• run /code_analysis all
• prepare Codex or Claude Code prompt packs
• import AI review outputs and regenerate final reports

Command Mapping

Run these commands from the repo root:

• python plugins/code-review-assistant/scripts/review.py /code-review-assistant:code_analysis <code-path>
• python plugins/code-review-assistant/scripts/review.py /code-review-assistant:add <tool> <report-path> --project-root <project-root>
• python plugins/code-review-assistant/scripts/review.py /code-review-assistant:add <tool> <report-path> --config <config-path>
• python plugins/code-review-assistant/scripts/review.py collect --config <config-path>
• python plugins/code-review-assistant/scripts/review.py first-pass --config <config-path>
• python plugins/code-review-assistant/scripts/review.py source-pass --config <config-path>
• python plugins/code-review-assistant/scripts/review.py blind-review --config <config-path>
• python plugins/code-review-assistant/scripts/review.py merge --config <config-path>
• python plugins/code-review-assistant/scripts/review.py export --config <config-path>
• python plugins/code-review-assistant/scripts/review.py all --config <config-path>

PowerShell wrapper:

• powershell -File plugins/code-review-assistant/scripts/review.ps1 /code-review-assistant:code_analysis <code-path>
• powershell -File plugins/code-review-assistant/scripts/review.ps1 /code-review-assistant:add <tool> <report-path> --project-root <project-root>
• powershell -File plugins/code-review-assistant/scripts/review.ps1 all --config <config-path>

Add Existing Reports

Use /code-review-assistant:add <tool> <report-path> when the static-analysis report already exists and should be imported instead of generated by the plugin.

The add command:

1. Copies the report into <output-dir>/raw-reports/<tool>/.
2. Creates or updates review.auto.yaml.
3. Enables the selected tool with mode: import.
4. Preserves existing tools already registered in the config.

Examples:

• python plugins/code-review-assistant/scripts/review.py /code-review-assistant:add cppcheck reports/cppcheck.xml --project-root sample_project
• python plugins/code-review-assistant/scripts/review.py /code-review-assistant:add coverity reports/coverity.json --config sample_project/code-review-assistant-output/review.auto.yaml

Path-First Flow

When the user supplies a code path directly, the wrapper should:

1. Create a subdirectory next to the target path.
2. Check whether canonical flawfinder report already exists at raw-reports/flawfinder/flawfinder.standard.txt.
3. If canonical report is missing, invoke internal $tool-flawfinder to generate:
   • raw-reports/flawfinder/flawfinder.standard.txt
   • raw-reports/flawfinder/flawfinder.detailed.txt
   • raw-reports/flawfinder/flawfinder.high-risk.txt
   • raw-reports/flawfinder/manifest.json
4. Run the wrapper with explicit canonical report wiring:
   • --flawfinder-report <canonical-path> --skip-flawfinder
5. Generate review.auto.yaml inside that subdirectory.
6. Execute collect, first-pass, source-pass, blind-review, merge, and export in order.
7. Keep all generated artifacts under that same subdirectory.

Hosted AI Flow

For Codex or Claude Code driven review:

1. Set ai.provider to codex or claude_code in the config.
2. Run the required stage or all.
3. Read prompt packs from review-output/ai-analysis/prompts/.
4. If external AI outputs exist, import them into:
   • review-output/input/ai/round1/
   • review-output/input/ai/source-recheck/
   • review-output/input/ai/blind-review/
5. Re-run the stage or all to merge imported outputs into the final report.

Output Expectations

• Tool outputs land in review-output/raw-reports/ and review-output/normalized/
• AI-stage outputs land in review-output/ai-analysis/
• Final reports land in review-output/final/

Prefer all unless the user explicitly wants a single stage.