The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's bootstrap mechanism in scripts/bootstrap.ts programmatically executes bun install to fetch Node.js dependencies like commander from the public NPM registry. This occurs at runtime within the skill's directory.
[COMMAND_EXECUTION]: To automate PR management and worktree cleanup, the skill executes a variety of shell commands including git, gh (GitHub CLI), du, and bun. These are triggered via the skill's playbooks and internal scripts.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, specifically pull request review comments processed by the babysit playbook and AI agent session transcripts analyzed by the worktree-audit.mjs script. The babysit playbook explicitly instructs the agent to treat this data as untrusted and never as instructions, which reduces the risk of accidental obedience.
[DYNAMIC_EXECUTION]: The skill includes several custom TypeScript and JavaScript utilities (e.g., watch-pr, orch, resume.mjs) that are executed by the agent to perform workflow automation. These scripts are invoked with user-supplied parameters through command-line interfaces.

poteto-mode