The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill instructs the agent to read and process external data, such as CI/CD logs via gh run view and maintainer feedback from Pull Request comments. While necessary for the skill's goal of unblocking PRs and handling failures, these inputs originate from outside the agent's immediate control and could potentially influence its subsequent actions.
Ingestion points: The agent reads data from GitHub CI/CD runs (gh run view) and developer feedback in markdown files (issue-comment.md, pr-comment.md).
Boundary markers: The skill does not currently specify clear delimiters or explicit instructions to disregard natural language commands that might be embedded in CI logs or maintainer comments.
Capability inventory: The agent has the ability to execute git commands (add, reset) and write files to the local file system using the write_file tool.
Sanitization: There are no specific sanitization or filtering steps mentioned for the content retrieved from external GitHub tools before it influences the agent's staging or patch generation logic.
[Local Command Execution]: The skill utilizes standard version control and platform CLI tools to manage the development lifecycle.
Evidence: Instructions include the use of git add, git reset, and gh run view to perform repository maintenance and check integration status.

prs