autonomous-backlog

Execute one deterministic autonomous backlog cycle for Taskrail tracked work.

Required Flow

1. Run go run ./cmd/taskrail validate.
2. Run go run ./cmd/taskrail next --json.
3. If no task is eligible, report that and stop.
4. Read the selected task file under planning/tasks/.
5. Run go run ./cmd/taskrail start <task-id>.
6. Implement in a TDD loop.
7. Run the appropriate test tiers.
8. Run manual testing using the autonomous-manual-test skill when the task changes visible Taskrail behavior.
9. Run go run ./cmd/taskrail verify <task-id> --result pass|fail --summary "...".
10. If additional work is discovered, create a follow-up task.
11. Finish as completed or blocked.

Rules

• never hand-edit planning/STATE.md frontmatter once the CLI exists
• never hand-edit task status fields once the CLI exists
• always keep evidence paths in notes and reports
• stop after one task