The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exposes a surface for indirect prompt injection by ingesting untrusted data from the repository's commit history.
Ingestion points: The git-stand.mjs script reads commit titles and author names from the remote repository via git log (found in templates/claude/hooks/git-stand.mjs).
Boundary markers: The ingested content is wrapped in triple-bracket fences (<<< and >>>) with an explicit warning to the agent that the content is data, not instructions.
Capability inventory: The skill enables a wide range of capabilities in .claude/settings.json, including file system operations, package management (npm), and repository management (git, gh).
Sanitization: The script sanitizes the input by removing any sequences of three or more < or > characters to prevent escaping the boundary markers.
[COMMAND_EXECUTION]: The skill installs a persistent hook and an extensive permission allowlist to facilitate automated workflows.
Evidence: The skill configures .claude/settings.json to allow numerous developer commands without prompting and registers git-stand.mjs as a SessionStart and UserPromptSubmit hook that executes automatically during the agent's session.

website-permissions