The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill implements a custom execution utility (run_block.py) that extracts and runs shell and Python code blocks embedded within the skill's own documentation (references/addressing-protocol.md). This facilitates the dynamic assembly and execution of scripts at runtime based on provided parameters.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface for indirect prompt injection because its primary function is to ingest and act upon untrusted external data (GitHub pull request review comments). The skill has extensive write capabilities, including committing code to the repository, pushing changes, and modifying pull request metadata via the GitHub API. While the process includes a verification phase using a separate agent context, the direct influence of external comments on agent behavior is a notable risk factor.
Ingestion points: GitHub pull request review comments (inline, general comments, and review bodies) fetched using the gh api blocks defined in addressing-protocol.md.
Boundary markers: The protocol defines specific structures for findings and replies (e.g., HTML trailers for stable IDs), but it does not specify explicit delimiters or 'ignore' instructions for the raw content of the review comments when they are processed by the agent.
Capability inventory: The skill can perform git commit, git push, modify PR titles/descriptions, post PR comments, and resolve/reopen review threads. It also executes shell commands and Python logic locally via run_block.py.
Sanitization: The skill relies on the agent's logical evaluation against repository standards and specifications rather than mechanical sanitization or filtering of the incoming comment text.
[COMMAND_EXECUTION]: The skill performs various shell operations using the gh CLI and git to manage pull requests and repository state. These commands are executed as part of the intended workflow to resolve review feedback.

resolve-review