The Agent Skills Directory

[SAFE]: The skill implements a structured workflow for GitHub issue triage. The instructions focus on process flow (gathering context, verification, recommendation, and outcome application) without including any executable scripts, network operations, or sensitive data access.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data (GitHub issue/PR bodies and comments) in SKILL.md (Step 1: Gather context). It mitigates potential instruction injection by requiring the agent to parse prior notes and compare requests against a known knowledge base (.out-of-scope/). All capabilities (labelling, commenting, closing issues) are restricted to the issue tracker context and triggered by human maintainer commands (/triage). Sanitization is performed through structured templates (e.g., 'Needs-info template' and 'Agent Brief').
[COMMAND_EXECUTION]: The skill mentions checking out PRs and running relevant tests in SKILL.md (Step 3: Verify the claim). This is a standard part of PR triage and is presented as a manual or safe tool-mediated verification step rather than arbitrary command execution. The openai.yaml configuration explicitly disables implicit invocation, ensuring actions only occur upon direct maintainer request.
[DATA_EXPOSURE]: The skill defines a standard practice for managing rejected feature requests using an .out-of-scope/ directory. This is a local knowledge base for the agent and does not involve exfiltrating data to external servers.

triage