The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches data from external sources that can be influenced by third parties, such as GitHub PR reviews, issue comments, and CI failure logs through gh pr view, gh api, and gh run view --log-failed. An attacker with repository access or the ability to influence CI logs could embed instructions to manipulate the agent. The skill lacks explicit boundary markers or instructions to treat this ingested content strictly as data.
[COMMAND_EXECUTION]: The skill relies on executing gh and git commands to manage pull request state, including committing changes and pushing code to the repository. These high-privilege operations are performed autonomously by the agent based on its analysis of the PR environment, which could be exploited if the agent is influenced by malicious external input.

iterate