The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands including gh, grep, and git to manage issue workflows, view issue details, and persist triage decisions in the repository. These operations are strictly scoped to the intended triage and documentation process.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from GitHub issue bodies, comments, and audit files. It successfully mitigates this risk by providing the agent with explicit instructions to ignore any instructions embedded within the processed data and to rely solely on its internal, evidence-grounded classification process. Ingestion points: Data is loaded via gh issue view and from docs/audits/ files. Boundary markers: The skill defines hard rules in its label and process documentation to prevent being influenced by user-submitted text. Capability inventory: The agent has the ability to write to local files, commit to the repository, and modify GitHub issue metadata. Sanitization: The skill specifically instructs the agent to use the --body-file flag when posting comments to GitHub to prevent shell injection risks associated with character escaping.

triage-issue