The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run across multiple scripts (fetch_issues.py, apply_triage.py, suggest_triage.py) to interact with the GitHub CLI (gh) and git. These operations are well-scoped to issue management tasks within the a2ui-project/a2ui repository and include standard error handling and validation. The use of the gh CLI for authentication ensures that the skill operates within the user's existing security context without managing raw credentials directly.
[DATA_EXPOSURE]: The skill writes temporary JSON data to the agent's scratch directory (<appDataDir>/brain/<conversation-id>/scratch/). This is standard practice for maintaining state across skill steps in this environment and does not expose sensitive user data to external parties.
[REMOTE_CODE_EXECUTION]: While the skill downloads issues from GitHub, it does not execute remote code or download untrusted scripts. The local dashboard (triage_dashboard.html) uses external libraries (Marked and DOMPurify) from a well-known CDN (JSDelivR) for rendering and sanitization, which is considered a safe practice for UI development.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub issue bodies and comments. However, it incorporates a mandatory human-in-the-loop step via a local web dashboard where a developer must approve or edit all AI suggestions before they are applied. This design effectively mitigates the risk of automated obedience to instructions embedded in issue text.

a2ui-issue-triage