The Agent Skills Directory

[COMMAND_EXECUTION]: Several internal Python scripts (e.g., scripts/batch_convert_templates.py, scripts/convert_to_csv.py) use subprocess.call() to invoke other Python scripts within the package for format conversion and parsing tasks.
[COMMAND_EXECUTION]: The README.md installation instructions reference external scripts (install-skills-mac.sh and install-skills-windows.ps1) that are not included in the skill definition. The Windows instruction specifically employs the -ExecutionPolicy Bypass flag to override default PowerShell security settings.
[PROMPT_INJECTION]: The skill processes untrusted external data from code diffs and PR descriptions, exposing it to potential indirect prompt injection attacks. 1. Ingestion points: Diff content and PR metadata ingested via prompt (SKILL.md). 2. Boundary markers: Guidelines instructing the agent to 'separate facts from assumptions' and 'stay in scope' (prompts/code-review.md). 3. Capability inventory: Ability to execute local scripts and write output files. 4. Sanitization: No explicit sanitization or filtering of analyzed code content is performed.

code-review