[dancinlife]

turn-close inject now checks each governance STORE for absence and emits one seed line per
missing one (was: the arch tree only):

• ① no ARCHITECTURE.json/.md at all (designSsot) → author the current-state tree (existing).
• ② ARCHITECTURE.md only → legs.arch is armed but legs.conv is not, so the 🧬 recurrence
  store (convergence.records[], JSON-only) stays silent-dead → ask to promote to
  ARCHITECTURE.json. ①② share the shipped archSeed knob.
• ③ no refs/heads/ing (ingBoardActive) → say the 🔄 ING leg is OFF and name the arming
  action (sidecar ing add). New opt-OUT knob ingSeed (default true, like archSeed).

WHY the ING seed: 1d627be assumed ING self-starts — store = a git ref ing add creates on the
spot, demand = engine-shipped commons ing-board. Only half holds. Zero creation friction is a
necessary condition, not a sufficient one: commons is generic, board-state-blind text, and
ing inject is silent with no items, so nothing ever says "THIS repo has no board and the 🔄 leg
is disarmed". The seed is that repo-specific "leg OFF + here is the arming action" signal.

CONDITIONAL imperative, deliberately: the line asks for a board only when the turn actually has
multi-step/handoff work, and says in so many words not to add an empty/fake item to silence it.
An unconditional "create a board" would let a one-shot repo buy silence with a fabricated board,
arming the Stop gate against fiction — the ARCH-PLACEHOLDER failure mode, with no lint to catch
it. There is no deterministic "this turn is multi-step" signal inside UserPromptSubmit (and Pi
passes a synthesized payload), so this is solved by wording, not by a firing condition.

No hook rewiring: turn-close inject already rides hooks.json + setup.ts + pi/sidecar.ts every
turn. Seeds are rendered whole or not at all (never truncated · inject-lint) and are dead
branches once the store exists (0B/turn).

QA (all pass): fresh repo → 2 seeds, no trio · ARCHITECTURE.md-only repo → trio(arch leg) +
promote seed + ING seed · fully-armed repo → 0B · archSeed:false/ingSeed:false silence each
independently · after sidecar ing add the ING seed disappears and the ref exists.

Design delegated to fable+sol. They split on whether an ING seed should exist at all (sol: commons
already covers it); taken fable's call, backed by the user's request and by commons being
board-state-blind. Sol's ARCHITECTURE.md-only gap is adopted as ②. Convergence turn-close-ts-1
updated in place — the "ING self-starts" clause was the wrong half of the lesson.

Follow-up (not in this PR): commons ing-board still says "repo-root ING.jsonl", pre-ref
wording that nudges the hand-written worktree file the seed warns against. Correcting it trips
COMMONS-OVERSIZED (config/commons.md is 11486B vs the 10300B cap, already over on main), so the
wording fix belongs with the commons trim it needs.

Co-Authored-By: Claude Opus 5 (1M context) noreply@anthropic.com