The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/pitch.py executes the official GitHub CLI tool (gh) via subprocess.run to retrieve repository metadata. The command is invoked using a list of arguments, which is a safe practice that prevents shell injection, and its usage is limited to read-only operations (gh repo view).
[EXTERNAL_DOWNLOADS]: The skill performs network requests to pypi.org and registry.npmjs.org using urllib.request.urlopen to fetch package summaries. These operations target well-known technology services and use hardcoded base URLs to ensure the requests are directed to legitimate registries.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from external sources, including README files and descriptions fetched from third-party repositories and package registries, which could contain untrusted input.
Ingestion points: Local README files and remote repository/package descriptions fetched via the drift command in scripts/pitch.py.
Boundary markers: The script does not wrap external data in security delimiters or warnings before returning it to the agent context.
Capability inventory: The skill's capabilities are restricted to read-only command execution (gh) and network reads from specific registries.
Sanitization: Content is normalized through whitespace removal and case folding, but the script does not perform specific sanitization for prompt injection patterns.

repo-pitch