The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub pull request comments and review feedback, which are provided by external contributors.
Ingestion points: The scripts/pr-review-loop.mjs script retrieves comments, reviews, and reviewThreads content via the GitHub GraphQL API.
Boundary markers: The skill instructions do not specify any boundary markers or delimiters to isolate the ingested data from the agent's instructions, which may lead the agent to follow instructions embedded in comments.
Capability inventory: The skill has significant write capabilities, including the ability to push commits to the repository, resolve review threads, and perform squash merges on pull requests.
Sanitization: No sanitization or filtering is applied to the ingested text before it is presented to the agent for classification and action.
[COMMAND_EXECUTION]: The skill performs repository operations using the GitHub CLI (gh) and Git, authenticated with a provided token.
The bundled script pr-review-loop.mjs uses execFileSync to run gh pr view and gh api graphql.
The SKILL.md instructions direct the agent to execute git push and gh pr merge commands.
All operations use the GITHUB_BOT_TOKEN environment variable for authorization.
The skill specifies a fixed path in the installation directory for its bundled script ($HOME/.agents/skills/...) to mitigate the risk of script shadowing attacks from the repository root.

gh-pr-codex-review-loop