The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill utilizes a workflow that ingests data from external GitHub Pull Request comments, which includes a section specifically titled "Prompt for AI Agent". This represents an indirect prompt injection surface where instructions embedded in external PR comments could attempt to influence the agent's code-editing behavior.
Ingestion points: Pull Request review comments fetched via the GitHub API (gh api) or WebFetch as described in the Workflow section of SKILL.md.
Boundary markers: The skill includes a specific filter to only process comments from authors whose username starts with sentry (e.g., sentry[bot]). It also defines a strict structure for parsing the comment body using specific markdown and HTML tags.
Capability inventory: The agent is authorized to use the Edit tool to modify repository source code and the gh CLI tool to interact with GitHub APIs.
Sanitization: The skill explicitly instructs the agent to "Verify First" by confirming the issue exists in the current code and evaluating the suggested fix against the root cause before implementation. It also includes a safety guideline to use AskUserQuestion if a fix is uncertain.
[COMMAND_EXECUTION]: The skill instructions include the use of the gh api command to retrieve review comments from GitHub. This is a legitimate use of the official GitHub CLI tool for the skill's primary purpose of repository maintenance and does not involve the execution of untrusted scripts or arbitrary commands.

sentry-code-review