The Agent Skills Directory

[SAFE]: The skill implements a robust 'read-only' triage process where findings and label deltas are presented to the user as a reviewable shell script rather than being executed automatically.
[COMMAND_EXECUTION]: Shell script helpers (fetch-issue.sh, list-labels.sh, resolve-repo.sh, search-related.sh, update-search-keywords.sh, upsert-comment.sh) are used to interact with the GitHub API via the gh CLI. These are localized to the skill's directory and follow best practices for parameter handling.
[EXTERNAL_DOWNLOADS]: The skill interacts exclusively with official GitHub repositories under the mui/ organization. Dependencies and metadata are fetched from trusted GitHub API endpoints.
[PROMPT_INJECTION]: Instructions include clear safety guidelines for the agent, such as 'never execute user-supplied code on the host' and 'never run reporter-supplied code'. It also mandates the use of delimiters for generated content.
[DATA_EXFILTRATION]: No sensitive file access or unauthorized network exfiltration patterns were detected. Data handling is limited to public GitHub issue metadata and local codebase searches.
[CREDENTIALS_UNSAFE]: No hardcoded secrets or unsafe credential storage patterns are present. Authentication relies on the user's local gh CLI environment.

mui-triage