A2UI GitHub Issue Triage Skill
This skill guides the process of fetching, analyzing, reviewing, and applying triage decisions to GitHub issues in the A2UI repository using a local interactive dashboard.
[!IMPORTANT] WRITING GUIDELINES When drafting replies, explanations, or any prose, refer to the natural-writing skill to ensure clarity, accuracy, and tone.
Workflow
Follow these steps to triage new or backlog issues:
Step 1: Fetch Untriaged Issues
Use the fetch script to retrieve all open issues lacking a priority label (P0, P1, P2, P3) in the A2UI repository.
- Run the fetch script to download the issues to a raw JSON file in your conversation-specific scratch directory:
python3 .agents/skills/a2ui-issue-triage/scripts/fetch_issues.py --repo "a2ui-project/a2ui" --output-file "<appDataDir>/brain/<conversation-id>/scratch/raw_issues.json"
Step 2: Analyze and Suggest Triage
Process the raw issues and generate recommended triage fields based on the project's triage criteria.
- Read the triage criteria reference document: triage_criteria.md and repository canonical templates: templates.md.
- For each issue in
raw_issues.json, evaluate its description and comments to determine:- Priority:
P0(Urgent),P1(High),P2(Medium),P3(Low), orNone.P0andP1require assignees. - Assignee: Recommended owner based on the affected component or area.
- Action:
investigate,assign_and_fix,needs_info,backlog,close_duplicate,close_invalid, orclose_resolved. - Labels: Applicable repository labels (e.g.
type: bug,component: lit renderer,status: first-line-handled,status: waiting-for-author-response). - Reply: A polite, structured draft response addressing the author using canonical templates.
- Priority:
- If an issue is a potential duplicate, perform at most three targeted GitHub searches to find matching canonical issues before suggesting
close_duplicate. - Natively Orchestrate Subagents: Instead of running a Python script to spawn subagents (which can fail due to local workstation gRPC credential policies), the parent agent should natively orchestrate the parallel evaluations:
- Load the first N issues (defaulting to 10, or as requested) from
raw_issues.json. - Call the
invoke_subagenttool in parallel for those issues. Prompt each subagent to analyze its assigned issue against the guidelines in triage_criteria.md and return a structured JSON block containingpriority,action,labels, andreply. - Once all subagents report back, compile their recommendations into the standard schema:
- Map component labels to suggested assignees using git log history if necessary, and include a short sentence in
assignee_reasonexplaining why they were chosen (e.g., "Suggesting gspencer because they recently modified related code"). - Inject
total_issues_count(preserving the total count from the raw issues JSON). - Save the final compiled payload to
issues_to_triage.jsonin the scratch directory.
- Map component labels to suggested assignees using git log history if necessary, and include a short sentence in
- Load the first N issues (defaulting to 10, or as requested) from
Step 3: Launch Review Dashboard
Launch the interactive web dashboard to allow the oncall engineer to review and refine the suggested triages.
- Start the local server as a background task, pointing it to your scratch directory:
python3 .agents/skills/a2ui-issue-triage/scripts/launch_dashboard.py --data-file "<appDataDir>/brain/<conversation-id>/scratch/issues_to_triage.json" --output-file "<appDataDir>/brain/<conversation-id>/scratch/triage_decisions.json"SetWaitMsBeforeAsyncto1000so the server runs in the background. - Iterative Batch Triage: The dashboard supports reviewing and applying issues in batches directly from the UI (
Apply Batch). As decisions are synced to GitHub in real time, the UI displays progress (X / Y updates made) and checkmarks next to items. The user can continue triaging remaining issues without closing the session. - Milestone Management: The dashboard displays milestone dropdowns and badges, enabling adding, changing, or removing issues from milestones.
- Wait for Completion: When finished, the user clicks "Finish & Exit" or "Abort". Once the user acts, the background task will complete, and you will receive a notification with the exit status.
- Verify Exit Status:
- If the task exited with status
0(approved), any remaining approved decisions are saved to the decisions file. - If the task exited with a non-zero status (abort), you MUST STOP and ask the user for further instructions.
- If the task exited with status
Step 4: Apply Decisions to GitHub
Once the dashboard task exits successfully, Execute Approved Decisions: Run the apply script to update the approved labels, assignees, and comments on GitHub:
python3 .agents/skills/a2ui-issue-triage/scripts/apply_triage.py --decisions-file "<appDataDir>/brain/<conversation-id>/scratch/triage_decisions.json"
Bundled Resources
- triage_criteria.md: Authoritative guide for classifying issues, assigning priorities, and drafting response messages.
scripts/fetch_issues.py: Script to query open, untriaged issues via the GitHub CLI.scripts/suggest_triage.py: Helper script to generate initial triage suggestions.scripts/launch_dashboard.py: Local HTTP server that opens the interactive web dashboard in the user's browser.assets/triage_dashboard.html: HTML/CSS/JS template for the triage review interface.scripts/apply_triage.py: Script to apply finalized triage decisions to GitHub.