forge-drivelisted

# forge-drive You own one issue from open to done, in one session. Nothing dispatches you again — there is no next job. When you stop, the issue stops. The cloud is your ledger, not your controller. It records what you declare and blocks exactly two things: deploy and close. Everything else is yours. ## The loop | Phase | You produce | Declared before starting | |---|---|---| | 1 · understand | a statement of what is wrong and how you reproduced it | `understand` | | 2 · plan | the files you will touch and why, and the branches you rejected | `plan` | | 3 · code | a working branch, build green, tests green | `code` | | 4 · self-review | your own diff read back against the acceptance criteria | `self-review` | | 5 · review | an independent verdict | `review` | | 6 · merge | the branch in the base it came from | `merge` | | 7 · ship | changelog line, close comment | `ship` | Declare each phase with `forge_phase` **before** you begin it (action `start`), and close it when it finishes (action `end`, with an outcome). That pair is your resume point: `resume_point` returns the newest phase you started and never ended, so a session that dies restarts there instead of at phase 1. A phase you never declared did not happen as far as any other session can see — including yours, after a crash. Phase 5 can send you back to phase 3. That loop has no counter — go around as many times as the verdict requires. Re-declare `code` each time so the journal shows the rounds. ## What this pr