The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a Bash script (scripts/find.sh) that executes standard system utilities such as curl, jq, unzip, and the GitHub CLI (gh).
[EXTERNAL_DOWNLOADS]: The script performs read-only network operations to skills.sh, clawhub.ai, and github.com to retrieve skill indices and documentation for search and vetting. These are recognized as legitimate service platforms.
[DATA_EXFILTRATION]: The skill performs local reconnaissance by scanning folder names within the ~/.agents/skills and ~/.claude/skills directories. While the purpose is to identify existing installations, this exposes local filesystem metadata.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface due to the ingestion and display of untrusted metadata (e.g., summaries and documentation) from third-party repositories.
Ingestion points: Metadata and raw SKILL.md content fetched from external registry APIs and GitHub repositories.
Boundary markers: External content is not wrapped in security delimiters in the output report.
Capability inventory: The skill has active network access and filesystem write capabilities for temporary data storage.
Sanitization: Input queries are URL-encoded via jq, but no specific sanitization is applied to third-party content processed by the agent.

find-skills