The Agent Skills Directory

[COMMAND_EXECUTION]: The bundled scripts/review.py executes local git commands to calculate file churn and repository paths. The implementation uses subprocess.run with argument lists and without a shell, which prevents command injection.
[PROMPT_INJECTION]: The skill processes data from the technical debt registry files, representing a surface for indirect prompt injection.
Ingestion points: Registry markdown files located in the docs/debt directory.
Boundary markers: The skill relies on agent instructions and a structured rubric rather than technical delimiters to prevent instruction override.
Capability inventory: Includes file system read/write access and local command execution via the git binary.
Sanitization: Content from the registry files is processed without explicit sanitization or filtering, but risks are mitigated by the agent's interactive rubric.
[DATA_EXFILTRATION]: The skill operates strictly on the local filesystem. No network-enabled tools or libraries are invoked, and no data is transmitted to external endpoints.

debt-ops-review