[Agentic] Improve the daily-test-improver with current findings

[marcelofukumoto]

Based on current finding when trying to improve what has been created by the workflow.
Added additional rules to the WORKFLOW prompt.
And added the QA/None tag as well since we were continuously adding.

One missing part now is the Milestone, I will do it on a different issue but want for it to verify what is the correct milestone it should be adding.

[github-actions]

🎯 Agentic Issue Triage

This is an internal improvement task for the daily-test-improver agentic workflow. The author documents findings from running the workflow and wants to improve the workflow prompt rules, refine QA tagging logic, and eventually handle Milestone assignment.

📋 Summary

Type: Task / Tech Debt
Area: Agentic workflows (CI/test automation)
Milestone: v2.15.0

The issue captures two improvements already made:

1. Added additional rules to the WORKFLOW prompt
2. Added QA/None tag support (to stop re-adding it unnecessarily)

One outstanding gap: Milestone assignment — the workflow doesn't yet determine and apply the correct milestone. This is deferred to a separate issue.

🔍 Related Issues

• #17452 - [Test Improver] Monthly Activity 2026-05 — active monthly tracking issue for the test improver
• #17364 / #17300 — past failure reports for the daily test improver (closed)

✅ Suggested Sub-tasks

• Add additional rules to WORKFLOW prompt
• Add QA/None tag handling to stop repeat application
• Implement Milestone detection and assignment logic (tracked separately)
• Validate updated prompt rules against recent test-improver runs
• Consider adding a dry-run/debug mode to the workflow for easier iteration

💡 Notes & Ideas

• When implementing milestone logic, the workflow will need to map issue context (e.g., target branch, existing milestone patterns like v2.15.0) to the correct milestone. Consider querying open milestones via the GitHub API and using heuristics or LLM reasoning to select.
• The QA/None fix is a good pattern — consider cataloguing all label categories that should only be applied once and adding guard logic for each.
• Relevant code likely lives in the workflow YAML or prompt definition files under .github/ or scripts/. Review those files to confirm the prompt changes are persisted correctly.

Generated by Agentic Triage for issue #17659 · ● 230.3K · ◷