Clarify broadcast-suppressed lane pickup fallback

Context

Night-shift coordination on 2026-05-20 surfaced a repeated workflow friction pattern: when the operator suppressed AGENT:* broadcasts because Gemini's harness was unstable, GPT and Claude treated that broadcast prohibition as if it blocked further implementation-lane pickup. That produced an avoidable idle window even though the repository had many non-conflicting open tickets and the operator had already made direct-DM coordination to Claude an acceptable substitute.

This is not a blame/shame issue. The useful signal is substrate friction: an existing rule was interpreted too literally because the skill payloads do not clearly distinguish "broadcast unavailable" from "work unavailable."

The Problem

The current substrate contains several correct anti-idle rules, but their interaction leaves a gap:

• #10970 codified post-review next-lane pickup.
• #11165 tightened post-review-pickup halt-state criteria around backlog self-survey and concrete context-exhaustion triggers.
• #11221 closed the "state intended lane, then idle" loophole.
• peer-role-mode.md still presents the canonical [lane-claim] example as to: 'AGENT:*'.
• post-review-pickup does not explicitly say that a temporary broadcast suppression requires a fallback coordination channel, not a halt-state.

When AGENT:* was operator-suppressed to avoid waking/crashing Gemini, the missing fallback language let agents over-escalate to a hold. The intended behavior was simpler: coordinate directly with the reachable peer (@neo-opus-ada in this incident), avoid Gemini, and either continue the active epic on non-conflicting lanes or widen to body/grid/docs/backlog tickets before declaring any halt.

The Architectural Reality

This is a skill/workflow-substrate issue, not a runtime bug. A narrow documentation/agent-contract patch should be enough.

Primary surfaces:

• .agents/skills/post-review-pickup/references/post-review-pickup-workflow.md — owns positive-ROI lane pickup and legitimate halt-state criteria.
• .agents/skills/peer-role/references/peer-role-mode.md — owns lane announce / [lane-claim] coordination shape and should name the direct-DM fallback when broadcast is operator-suppressed.
• Optionally AGENTS.md §critical_gates / §swarm_topology_anchor only if the skill-level fix cannot make the rule interaction discoverable enough. Prefer skill-level tightening first to avoid turn-loaded bloat.

Empirical duplicate sweep before filing:

• ask_knowledge_base(type:'ticket', query:'duplicate tickets broadcast suppression direct DM fallback halt-state watchdog post-review-pickup lane selection') surfaced related closed tickets #10970, #11165, #11221, #10668, #11038, but no direct duplicate for broadcast-suppressed lane pickup fallback.
• Local grep over resources/content/issues, resources/content/discussions, and .agents/skills found the same related lineage plus [lane-intent] / [lane-claim] semantics, but no explicit fallback contract for operator-suppressed broadcasts.
• Live gh issue list searches for halt-state backlog self-survey broadcast direct DM idling, nightshift watchdog idle halt post-review-pickup, broadcast suppression fallback direct DM coordination, and Gemini harness crash broadcast direct DM returned no open duplicates. One broad search hit #11628, which is an unrelated KB ingestion operations epic.

The Fix

Clarify the lane-pickup contract:

1. AGENT:* broadcast is the canonical visibility path, but it is not the only collision-prevention primitive.
2. If the operator explicitly suppresses broadcasts because a peer harness is unstable, agents must use the operator-authorized reachable peer DM as the coordination substitute.
3. Broadcast suppression alone is not a legitimate halt-state.
4. Before declaring halt, agents must widen the lane search beyond the immediately active epic to non-conflicting open backlog work such as body, grid, documentation, testing, or other Project 12 lanes.
5. Night-shift/watchdog pulses must not degrade into "wake -> ack -> nothing to do" when positive-ROI backlog lanes exist.

Contract Ledger Matrix

Acceptance Criteria

• post-review-pickup-workflow.md clarifies that broadcast suppression is not a halt-state and requires an operator-authorized direct-DM fallback when available.
• post-review-pickup-workflow.md requires broad positive-ROI lane discovery before halt, explicitly including body/grid/docs/general backlog when the active epic is gated.
• peer-role-mode.md keeps AGENT:* as canonical [lane-claim] visibility but documents direct-DM fallback semantics for operator-suppressed broadcasts, including that the fallback is scoped to the named reachable peer and must not wake the suppressed peer.
• The anti-pattern table gains an entry for "broadcast unavailable => idle" / "wake -> ack -> nothing to do".
• Verification evidence demonstrates the updated text references #10970, #11165, and #11221 rather than duplicating or contradicting them.
• Cross-family review per pull-request workflow; human-only merge boundary preserved.

Out of Scope

• Runtime wake-subscription changes.
• Re-enabling broadcasts to Gemini while the operator's no-broadcast constraint is active.
• Changing the core [lane-claim] authority semantics from #11536/#11537.
• Mandating that every watchdog pulse must start code work; the rule is to search honestly and use the available coordination channel, not to manufacture unsafe work.

Avoided Traps / Gold Standards Rejected

• Blame/shame framing: rejected. Shame hides friction; the durable value is a clearer substrate contract.
• Treat direct-DM fallback as universal replacement for broadcast: rejected. Broadcast remains canonical; direct-DM fallback applies when the operator suppresses broadcast and names/permits a reachable peer channel.
• Over-broaden into a new coordination system: rejected. Existing post-review-pickup, peer-role, and A2A direct messages are sufficient.
• Idle because the current epic is gated: rejected. The backlog is broad; non-conflicting body/grid/docs/general tickets are legitimate lanes.

Related

• #10970 — Codify post-review immediate next-phase pickup discipline
• #11165 — Revise post-review-pickup halt-state to require backlog self-survey
• #11221 — Close §15.6 intended-next-lane loophole
• #11536 / #11537 — [lane-intent] / [lane-claim] authority semantics

Handoff Retrieval Hints

• query_raw_memories(query="broadcast suppression direct DM fallback halt-state nightshift watchdog idle")
• query_summaries(query="wake ack nothing to do post-review-pickup direct DM Claude Gemini crash")
• File anchors: .agents/skills/post-review-pickup/references/post-review-pickup-workflow.md, .agents/skills/peer-role/references/peer-role-mode.md