ticket locks are the format's .lock.md claim; PENDING placeholders + 6h timer removed (fix #1420)

[suleimansh]

Implements the plan agreed on #1420 (comment): the code now speaks the same lock the prompts define, and the old mechanism is gone.

What changed

spike-locks.ts → ticket-locks.ts. The claim is now one tickets/<STEM>.lock.md per ticket holding CLAIMED: <AGENT_ID> — the file ticketing_format.md defines — instead of PENDING: placeholders pre-filling .spike.md/.plan.md. The daemon still writes, commits, and pushes the locks before agents start (an agent only pushes at session end onto its own branch, so it can never place the lock in time; cloud runs can't push at all — #1320).

The 6h staleness release is deleted (releaseStaleSpikeLocks, SPIKE_LOCK_STALE_MS), per the issue: a coordinator can hold a ticket for days, and a timer releasing under a live agent re-opens the double-work window. Release is now explicit:

• the fanned-out agent's pinned prompt tells it to delete the lock in the same commit as its plan — the merge that lands the work frees the ticket;
• a human can lift a dead agent's claim with the new Release lock button on the ticket's detail page (sendReleaseTicketLock: delete, commit, best-effort push — a release only one machine sees would leave the ticket claimed where it matters).

Readers switched over. readTickets/readTicket mark locked from the .lock.md sibling's existence (plus lockedBy from its CLAIMED: line, shown as the badge/button tooltip); the lock file never renders as a ticket row; the daemon's spikeCandidates filter is unchanged in shape but no longer runs a stale-lock sweep. Tickets wear a "claimed" badge in the list and on the detail page.

Existence is meaning again: a .plan.md that exists is a plan (the placeholder ambiguity is gone), which also un-breaks the planned-quick-wins fail-closed path for tickets currently being planned.

Not in this PR (flagged for follow-up)

• planned-quick-wins.ts still parses the pre-Improve lock mechanism #1420 Effort: quick-win / Consensus: keys, so new-format plans never auto-queue — next slice.
• Implementation runs (queue drains) still claim via A pinned queue entry can be handed out twice across ticks (web target, daemon restart) #1253/Queue claims are per-machine: two daemons draining the same queue double-assign an entry #1313 entry claims rather than .lock.md; unifying those into the whole-life lock is the issue's larger vision.

Tests

• the-framework: 1650 tests, 0 fail (lock module rewritten: batch acquire, race skip, commit rollback, push tolerance, feature-branch guard; new release paths incl. a real-git end-to-end RPC test)
• framework-dashboard: 671 tests, 0 fail (claimed badge, release button flow, failure surfacing)

🤖 Generated with Claude Code