The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from Pull Requests. It includes strong architectural mitigations, including explicit instructions to treat PR descriptions as untrusted and to ignore any embedded commands or instructions. It also commands the agent to escape repository-controlled labels in any diagrams or Markdown output to prevent XSS. This behavior is validated by the 'hostile_and_large_input' eval cases which include prompt injection and XSS payloads.
[SAFE]: The skill adheres to the principle of least privilege, explicitly forbidding the execution of target code, builds, or any mutating operations like pushes or merges. The included Python test utility (provider_stub.py) is implemented using security best practices, such as command normalization and restricted file permissions for local state handling.

pr-explainer