The specs catch up with today's code: branch lifecycle, routine locks, bridged gates

[brillout]

Follow-up to #1655: five code PRs merged today (#1653, #1656, #1658, #1662, #1664) changed behavior the fresh specs describe, and none of them updated per-file specs — the spec install had not landed while they were developed. This PR brings the specs back in line with the code as it stands on main, per https://raw.githubusercontent.com/brillout/sdd/refs/heads/main/sdd.md.

What tracks what:

• An agent that committed nothing leaves no branch behind (#1650) #1653 / A branches/ directory that is not a git worktree is left alone, never acted on as the main checkout (#1654) #1656 / The run-id branch an agent branched away from goes with its checkout (#1657) #1658 (worktree and branch lifecycle) — worktrees, merged-worktrees, store/worktree, branch-names, branch-links, daemon-runtime specs (and their test specs) now describe: an agent that committed nothing loses its branch with its checkout, a branches/ directory that is not a git worktree root is refused before any git runs, and the birth branch (tf-agent-) goes with the checkout when the kept branch contains it.
• One triage at a time through a routine lock on tf-data, not a pinned branch (#1659) #1662 (routine locks) — stale-branch.SPEC.md and its test spec are deleted with their module; new routine-locks.SPEC.md / routine-locks.test.SPEC.md describe the routines/.lock.md guard on the data branch; auto-pm, daemon-services, RoutineWork, the triage preset specs and the affected test specs now describe the lock instead of the pinned-branch abort; the root and src/ roll-ups drop the retired stale-branch release.
• A cloud session asks its questions when the bridge is on, and they are answered like a local gate (#1554) #1664 + No decide-alone mode for a cloud session: its gates are a local agent's gates (#1554) #1667 (bridged gates) — new bridge-question.SPEC.md; the bridge, open-questions, agent-framing, system-prompt, turn-gate, cloud-driver, Chrome-extension and dashboard-component specs now describe a cloud session's questions surfacing through the Claude web bridge and being answered like local gates. Per No decide-alone mode for a cloud session: its gates are a local agent's gates (#1554) #1667 there is no decide-alone mode: the protocol set is back to four, the hands-off protocol says only "land everything", and a cloud session's gates are a local agent's gates (unattended.SPEC.md was dropped with its prompt).
• Merged main through A web run's row says what its cloud side is doing, not "in cloud" forever (#1668) #1669, whose own spec updates came with it; the one overlap (client.SPEC.md) is resolved by combining both sides.

Specs checked and found still accurate were left untouched. FEATURES-SPEC.md remains untouched (the merged PRs updated it themselves).

🤖 Generated with Claude Code

https://claude.ai/code/session_01CSM4FVUbnP4PJJ9sawMfLb

[suleimansh]

Heads-up: #1667 (merged) removed unattended.md as you asked on #1664, so prompts/protocols/unattended.SPEC.md and the "bridge off → the session decides alone" sentences here are already stale. Drop them from this PR, or tell me and I'll push the fix onto your branch.

[brillout]

Done in fc2053b: merged main (bringing #1667 in), deleted prompts/protocols/unattended.SPEC.md, and realigned every spec that described the decide-alone split — turn-gate, system-prompt(+test), agent(+test), cli, protocols/SPEC.md, hands_off, and the extension's SPEC.md — to the current contract: four protocols, hands-off framing says only "land everything", and a cloud session's gates are a local agent's gates. One deliberate mention of the removed mode survives as a Rationale note in turn-gate.SPEC.md, explaining why the hands-off protocol says nothing about gates.

---

Generated by Claude Code