The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to monitor and respond to GitHub pull request comments and review feedback. This creates a surface where external input (from PR comments) can influence the agent's actions, such as performing code changes or merging the PR. The skill includes instructions to the agent to prioritize the user's original intent and provides structured ways to handle disagreements, which serve as manual verification steps. The following evidence defines the surface:
Ingestion points: The skill utilizes the gh API to fetch issue comments, review comments, and review states in both SKILL.md and the land_watch.py script.
Boundary markers: The skill instructs the agent to prefix its comments with [codex] and identifies specific keywords like ## Codex Review to distinguish automated feedback.
Capability inventory: The agent has the capability to perform git commit, git push, and gh pr merge, as well as make decisions about merging based on the fetched text.
Sanitization: The utility script includes a function to remove control characters from terminal output, and the markdown instructions provide a "Context guard" to prevent unintended changes.
Command Execution: The skill utilizes standard command-line tools, including git and gh (GitHub CLI), and executes a provided local Python script (land_watch.py). These tools are used for their intended purpose of interacting with the repository and the GitHub API to automate the merging process.

land