The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes external processes for validation tasks.
The scripts/review_skill.py script uses subprocess.run to execute a sibling script, quick_validate.py, for structural verification.
The tests/test_review_skill.py file uses subprocess.run to execute the reviewer script during testing.
These executions are restricted to local sibling scripts within the tool's own suite.
[DYNAMIC_EXECUTION]: The reviewer script dynamically loads modules at runtime.
scripts/review_skill.py uses importlib.util to load audit_skill_delivery.py from a sibling skill-governance directory.
This dynamic loading is used to integrate modular components of the broader skill-creation suite and does not load user-controlled or remote code.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and evaluate third-party skill content, creating an attack surface for indirect prompt injection.
The skill implements robust mitigations, including documentation that explicitly instructs the agent to treat target content as untrusted evidence rather than instructions.
The reviewer scripts perform static analysis only (e.g., using Python's ast module for syntax checking) and do not execute the target skill's code or dependencies.
[DATA_EXPOSURE]: The skill contains automated scanners designed to detect sensitive information such as API keys and personal paths.
The detection logic in scripts/quality_review.py and scripts/review_skill.py is implemented using regular expressions for known secret patterns (e.g., sk- tokens).
The implementation is safe as it identifies the presence and location of potential secrets without capturing or exposing the actual credential values in the output reports.

skill-reviewer