A Starko AI run used to end at "here is a pull request". It now ends with proof, and you decide how much of the loop runs on its own.
Task kinds, sizes and budgets
Every ticket a run picks up is classified as a bug or a feature and sized Small, Medium or Large. The size sets the token budget and whether the reviewer agent is skipped for trivial changes. Kind and size are mirrored to workspace tags, and a ticket you tag yourself is treated as pre-classified.
Independent verification
After the agent finishes, Starko runs the project's own test command in the sandbox and records the real exit code. That result goes on the Evidence tab, into the PR body and into the reviewer agent's prompt. If a continuation opens a new pull request, the old one is marked superseded rather than left hanging.
Evidence screenshots now follow GitHub's conventions: PNG, light theme, element crops with context, a highlight stroke around the change, personal data masked, descriptive file names.
Autonomy policy per project
Each project decides what Starko AI may do without asking: open pull requests only, auto-dispatch small fixes, or ask before every run. A reviewer revise or a red CI re-dispatches the run within a per-project cap. CI state rolls up across check suites, and a revise on a stale head is skipped.
Project settings and secrets
A gear next to the repository opens the project's Starko AI settings: instructions, autonomy, the commands to run, and a Vercel-style environment editor. Paste a block of KEY=value lines and it splits into rows; keys are shown masked and merged on save. The agent is told which variable names exist, never their values.
Ledger and billing
The Starko AI settings page shows what each merged pull request bills at list price next to its platform cost. Runs meter at 2x tokens instead of 4x.
Reliability
A run whose sandbox cannot launch fails in seconds with the real cause on the ledger instead of a 45-minute timeout. Continuations resume the last finished run, so an errored run can no longer swallow your feedback.