The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to ingest and process text provided by other 'source skills' via the public_body and debug_body fields. This content is then assembled and posted to GitHub. This pattern presents a surface for indirect prompt injection where instructions embedded in the source text could be propagated to GitHub. The skill includes basic sanitization by stripping JSON fences and requiring ASCII-only output, which helps mitigate accidental schema confusion.
Authorized Command Execution: The skill utilizes the GitHub CLI (gh) to interact with issues and pull requests. These operations are restricted to the microsoft/apm repository and are limited to commenting and patching actions. The skill explicitly forbids high-privilege operations such as merging, labeling, or assigning reviewers, adhering to the principle of least privilege for its intended use case.

autopilot-comment