Repository navigation
v5.28.0
Minor release: a headless verify FAIL builds its own fix phase before reporting. In one project using headless runs, 11 of 14 unattended features in the ten days to 2026-10-19 ended at least one run verify-fail. Each one sat idle until the caller's next hourly poll re-ran /ship:go to build the fix phase the run had just written. That came to 17.6 hours of waiting against 27.3 hours of work.
Changed
/ship:go --headlessruns bounded fix rounds (new section 6b in the go skill). After a FAIL it rebuilds the pending fix phase and verifies again, up to--fix-rounds {n}times (default 2,0restores the old behavior). Every round after the first runs only while the failing set shrinks: FAIL criteria, non-lowopen bugs, FAIL checklist rows and a FAIL trial merge. A stop, a null verdict, or a FAIL with no fix task ends the loop.verify-failis reported only after that, and itsdetailnames the rounds used and the failing-set trend. Interactive runs are unchanged.ship/docs/headless.md§1, §3 and §8 now describe the fix rounds instead of "go never auto-retries".