The Agent Skills Directory

[SAFE]: The skill performs local static analysis of source code to identify repeated utility class strings. It does not perform network operations or access sensitive system paths.
[COMMAND_EXECUTION]: The scripts/check_class_extraction.py script uses subprocess.run to call git for repository metadata and staged file lists. The commands are constructed using internal logic and do not interpolate untrusted external data into the shell execution path.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted source code from the project environment. The evidence chain includes ingestion from project files (Ingestion points: scripts/check_class_extraction.py), a lack of explicit boundary markers in the report, a capability inventory limited to local file system writes and git metadata retrieval, and sanitization based on regular expression filtering (_looks_like_classes). While this presents an attack surface for indirect prompt injection, the skill's functionality is limited to reporting findings back to the agent or user, which does not provide a mechanism for arbitrary code execution or significant privilege escalation.

class-extraction-guard