The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes GitHub pull request comments, review bodies, and discussion threads, which are untrusted external data sources that could be used to influence agent behavior.
Ingestion points: External data is ingested via the scripts/get-pr-comments helper, which fetches all relevant review content from the GitHub API.
Boundary markers: The SKILL.md and references/agents/pr-comment-resolver.md files provide strong instructions to treat comment text as untrusted and to never execute shell snippets or commands found within them.
Capability inventory: The skill has access to the Bash tool, file system modification capabilities, test execution tools (pytest, bun test, rspec), and Git repository management (git commit, git push).
Sanitization: The skill relies on natural language instructions for the LLM to handle untrusted input safely rather than using programmatic sanitization or filtering.
[DYNAMIC_EXECUTION]: The workflow requires the agent to implement code fixes and subsequently run "targeted tests" on those modifications. This involves the execution of code that has been dynamically generated or altered by the agent at runtime.
[COMMAND_EXECUTION]: The skill extensively utilizes the GitHub CLI (gh) and standard shell utilities like git, bash, and jq to interact with remote repositories and process data.
[PRIVILEGE_ESCALATION]: The skill is authorized to perform git commit and git push operations. This allows for permanent modifications to the target codebase, a high-privilege capability that is triggered based on the evaluation of external pull request feedback.

resolve-pr-feedback