The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/app_reviews.py modifies the Python search path (sys.path) at runtime to include a directory (../../copycat/scripts) located outside of the skill's own directory tree. This allows the skill to load and execute code from arbitrary locations in the parent file structure if specific folder names are present.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The file scripts/copycat_reviews/google_play.py imports functionality from a package named copycat_store.google_protocol. This package is neither a standard library nor provided within the skill files, indicating a dependency on external code that cannot be verified for safety.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted content in the form of public app reviews from the Apple App Store and Google Play Store. Malicious instructions embedded in these reviews could potentially influence the AI agent's behavior during the analysis phase.
Ingestion points: Untrusted review text is fetched in apple_store.py and google_play.py and stored in reviews.jsonl and evidence.md files.
Boundary markers: The skill does not use clear boundary markers or system instructions to delineate untrusted review content from the agent's primary instructions.
Capability inventory: The skill enables the agent to execute Python scripts that perform network requests and file system operations.
Sanitization: While the scripts perform basic text formatting, they do not implement any sanitization or filtering to detect or neutralize prompt injection attempts within the review data.

copycat-reviews