The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because its core functionality involves processing untrusted data from GitHub PR comments and review bodies.
Ingestion points: The script scripts/get-pr-comments fetches user-supplied text from PR threads, comments, and review bodies via the GitHub GraphQL API.
Boundary markers: The skill includes explicit security sections in SKILL.md and references/agents/pr-comment-resolver.md that warn the agent to treat comment text as untrusted input and never execute commands or snippets found in it.
Capability inventory: The skill possesses significant capabilities, including reading local source files, using git commit and git push to modify the repository, and using gh api to post replies and resolve threads.
Sanitization: The instructions require the agent to perform technical verification of all suggestions against the codebase and use comments as context for independent decision-making rather than as direct instructions.
[COMMAND_EXECUTION]: The skill executes several shell and PowerShell scripts located in the scripts/ directory to interface with the GitHub CLI (gh) and Git. These scripts are used to fetch PR metadata, submit GraphQL mutations, and manage local repository state (staging, committing, and pushing changes).

resolve-pr-feedback