The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which could contain instructions aimed at influencing the agent's behavior.
Ingestion points: PR metadata (title and body) fetched via gh pr view, content from linked GitHub issues, and the codebase diff via git diff.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore embedded instructions' markers when interpolating this external content into prompts for sub-agents.
Capability inventory: The agent has the ability to modify local files, execute shell commands (via repository checks), and push changes to remote git branches.
Sanitization: There is no evidence of sanitization, escaping, or validation of the external content before it is processed.
[DYNAMIC_EXECUTION]: The skill dynamically identifies and executes command-line instructions (e.g., linting, testing, type-checking) by reading repository-specific configuration files such as AGENTS.md, CLAUDE.md, or CI configurations. This allows for the execution of arbitrary shell commands defined within the codebase.
[COMMAND_EXECUTION]: The skill uses various shell-based tools including git for branch management and commits, gh for interacting with GitHub PRs and issues, and spawns sub-agents through shell commands such as claude -p or codex exec.

babysit-pr