Fix runs got memory, context and a way to talk to you.
Warm sandboxes and a repo map
Each project keeps a warm sandbox snapshot with dependencies installed, so a run boots in seconds and always fetches the latest commit. A "Learn project" research run writes a repo map that later runs read on demand, instead of rediscovering the codebase every time.
Learnings, concerns, deviations
Runs hand off with a contract: LEARNING lines become backlog tickets for the project and are staged back into the sandbox as a file for the next run; CONCERN and DEVIATION lines are shown on the run so a reviewer knows what the agent was unsure about.
Clarifying questions
When a run needs something only you know, it @mentions the ticket owner and collaborators (assigning first if nobody is), and a plain reply on the ticket resumes it. If nobody could be assigned, the person who invoked the run is tagged.
Stop, and see state on the board
Live runs have a Stop button. A ticket moves to In Progress when a run starts, and the board shows a spinner on tickets Starko AI is working on. Evidence keeps the run's last commands and every round's screenshots, and the newer GitHub token format is scrubbed from transcripts.
Eval replay
Flawed cases can be re-run against the current prompt and compared, so a prompt change is measured before it ships.
Also in this release
- Ticket files (attachments and files embedded in the description) are staged into the fix sandbox.
- Photo and file attachments on ticket comments; an Attachments section below Relations on the ticket.
- Operators can delete a WhatsApp template from Starko.
- The inbox shows "Starko replied" on AI-handled threads where the customer owes a reply.
- Date filters are built in your timezone, day by day.
- Admins manage every role below owner; invite controls are hidden from members.