When a project is linked to a code repository, Starko AI can already turn a ticket into a draft pull request. This release wraps a full quality loop around that process:
Tickets get triaged before a run is spent
New tickets in repo-linked projects are automatically classified — bug, feature request, docs gap, or question — and under-specified bug reports get an internal pre-flight note listing exactly what's missing. You see what to ask the customer before an agent run is spent on incomplete information.
Every draft PR gets a second opinion
After the agent ships a draft pull request, a reviewer pass reads the full diff and posts a risk-rated internal assessment (green, yellow, or red) so you know what you're merging before you open it. If the review finds concrete defects on a first attempt, the agent gets exactly one revision round to address them — bounded, so a fix can't loop forever.
It asks instead of guessing
When a ticket is genuinely ambiguous, the agent can now end its run with one specific clarification question instead of guessing or failing. The question appears as an internal comment; reply with @starko and your answer, and the run resumes on the same branch with the context it needed.
Give the agent standing instructions
Project settings gain an agent instructions field — conventions, gotchas, "always run the tests in package X" — that is injected into every fix run for that project. Think of it as a per-project briefing document the agent reads before touching your code.
Outcomes are tracked
Every run now records how it actually ended — merged, closed without merging, blocked, or needing more information — so the loop learns what worked. This feeds future improvements to the agent.
Live for all workspaces with a linked repository — nothing to configure except the optional project instructions.