The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses gh (GitHub CLI) to interact with pull requests and fetch review threads via GraphQL. It also uses standard git commands for branch management, commits, and PR creation. These operations are scoped to the current repository and performed within the context of developer-initiated workflows.
[INDIRECT_PROMPT_INJECTION]: The skill explicitly acknowledges the risk of malicious instructions embedded in review comments (described as 'untrusted input'). It implements several mitigation strategies: treating 'Prompt for AI Agents' sections as reports rather than instructions, requiring mandatory human approval for every code change, sanitizing reviewer guidance before display, and prohibiting the use of fetched text as direct shell input. The severity is assessed as LOW because these are vulnerability surfaces (ingesting external data) with robust internal guardrails.
[REMOTE_CODE_EXECUTION]: While the skill fetches data from GitHub's API, it does not pipe remote content to a shell or execute downloaded scripts. All code modifications are performed locally using the Edit tool after user validation.
[DATA_EXFILTRATION]: The skill's network operations are restricted to authenticated GitHub API calls (gh api) for the specific purpose of the skill. It contains explicit instructions to never read sensitive local files like .env, SSH keys, or credentials.

autofix