The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it is designed to ingest and analyze untrusted source code provided by users.
Ingestion points: Source code files provided by the user are read by the agent and analyzed by the pre-review.py script.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' warnings when the agent processes the user's code content, which could allow malicious instructions in comments to influence behavior.
Capability inventory: The agent has the capability to read local files and execute the pre-review.py helper script.
Sanitization: No sanitization or filtering of the input code is performed before analysis to mitigate potential prompt injection attempts.
[COMMAND_EXECUTION]: The scripts/pre-review.py file executes an external command as part of its analysis workflow.
Evidence: The script uses subprocess.run to call the ruff linter on the target file. While this is an intended feature of the skill to provide automated feedback, it involves executing a CLI tool from a script included in the skill package.

clean-code-reviewer