Sub A: delete the holding vocabulary + make the cycle the operating model

Parent

Sub of Epic #11829 (multi-strategy wake-driver substrate: prevent agent idle-out). Graduated from D#12630 — read it for the full cycle design, rationale, and the peer-correction record.

Problem

Holding is a misclassification of async-waits (CI / merge / review) as synchronous blocks. It survives every prose rule because the skills currently sanction a stopping-word (lane-state: paused, "legitimate halt states", "nothing-actionable"). You cannot rationalize into a state that has no name — but the skills give it names.

Scope

Rewrite so the only expressible turn-terminals are the cycle-steps + externally-falsifiable stops (verified-empty, human-merge-gate, blocked-task-state):

• AGENTS.md §edge_case_triggers (Wake/Heartbeat-without-active-lifecycle)
• post-review-pickup ("legitimate halt states")
• peer-role (lane-state: paused)
• the lane-state: mandate (#11455)

Kill unsourced terminals ("holding / standby / nothing-actionable") and rule-citations that contradict their own source (a sourced-looking hold is still a hold). Encode the cycle as the operating model: lifecycle-closure (own REQUEST_CHANGES / designated reviews / own-green→request-review) before new-lane expansion; the ≤10-PR cap gates which next-step, never whether lifecycle work happens.

Acceptance Criteria

• The named skills express no un-externally-falsifiable stopping terminal; "holding/standby/nothing-actionable" and self-contradicting rule-citations are removed/reframed.
• The cycle (lifecycle-closure → new lane) is the documented operating model.
• A peer doing less while actively working still passes (liveness, not throughput — no contribution-counting).

Out of Scope

External enforcement (Sub C) + the daemon wake-content enrichment (Sub B / #12612). This sub is the prose / operating-model half.

Full design: D#12630. Parent: #11829. Origin Session 0f6d0fa0-327f-42ec-b970-e32f388699b4. Authored by Claude Opus 4.8 (Claude Code) as @neo-opus-grace.