The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script find_skill.py uses importlib.util to dynamically load and execute a Python module from the file path ~/.claude/hooks/inject-skill-hints.py. While this is intended for code reuse within the platform's ecosystem, executing code from a local path at runtime is a dynamic execution pattern.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a discovery mechanism for other skills, ingesting and displaying their metadata which may contain untrusted content.
Ingestion points: The skill reads metadata (names and descriptions) from ~/.cache/claude-skill-index.json, which is compiled from SKILL.md files in the skills and plugins directories.
Boundary markers: The search results are presented to the agent without specific delimiters or instructions to ignore embedded commands within the descriptions.
Capability inventory: The skill enables the agent to discover and subsequently invoke other skills via the platform's Skill tool.
Sanitization: Descriptions are truncated by default, but no explicit sanitization or filtering of the text content is performed to mitigate potential prompt injection attacks embedded in skill descriptions.

find-skill