The Agent Skills Directory

[PROMPT_INJECTION]: The maintenance script scripts/fetch-and-transform-hooks-docs.sh implements an automated documentation refresh pipeline that is susceptible to indirect prompt injection.
Ingestion points: The script fetches content from several external documentation URLs (e.g., code.claude.com, docs.github.com) into /tmp before processing them.
Boundary markers: The LLM prompt used to transform the documentation (rwr:doc-to-skill) lacks delimiters or explicit instructions to ignore instructions embedded within the source text.
Capability inventory: The skill utilizes the claude CLI for content transformation and sed for updating the local references/platform-coverage.md file.
Sanitization: No validation or sanitization of the fetched content is performed beyond a minimum file size check.
[EXTERNAL_DOWNLOADS]: The scripts/fetch-and-transform-hooks-docs.sh script is configured to download content from multiple external domains.
The script fetches documentation from official sources including code.claude.com, docs.github.com, docs.cursor.com, docs.windsurf.com, and ampcode.com.
These are well-known technology and documentation services, and the network activity is tied to the skill's intended maintenance functionality.
[COMMAND_EXECUTION]: The skill executes shell commands for maintenance and provides various examples of command execution in its technical guides.
The script scripts/fetch-and-transform-hooks-docs.sh executes curl, claude, and sed to update local files.
Educational reference files such as references/hooks-cjs.md and references/hooks-python.md demonstrate the use of execFileSync and subprocess.run for executing local development tools like ruff, git, and prettier.

hooks-guide