The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection via its documentation refresh script.
Ingestion points: The scripts/fetch-and-transform-hooks-docs.sh script fetches content from external documentation URLs (including code.claude.com, docs.github.com, and ampcode.com) using curl.
Boundary markers: The script passes the downloaded content directly into a claude -p prompt without using delimiters or instructions to ignore embedded commands.
Capability inventory: The AI model processes the external data and the script then writes the output to local reference files within the skill directory.
Sanitization: No sanitization, validation, or filtering is performed on the downloaded content before it is interpreted by the AI model.
[EXTERNAL_DOWNLOADS]: The documentation refresh script performs network operations to fetch data from external sources.
Evidence: curl --silent --fail --max-time 30 --output "$tmp_file" "$url" in scripts/fetch-and-transform-hooks-docs.sh.
Sources: The script targets official documentation domains for Claude Code, GitHub Copilot, Cursor, Windsurf, and Amp.
[COMMAND_EXECUTION]: The skill provides a shell script intended for manual execution that performs network requests and invokes AI command-line tools.
Evidence: scripts/fetch-and-transform-hooks-docs.sh contains commands for curl, claude (AI CLI), and sed to modify local files.

hooks-guide