Add AGENTS.md rule 7 pointer for broadcast fallback

Context

PR #11670 resolved the immediate skill-atlas gap from #11669 by documenting that operator-suppressed AGENT:* broadcasts are a coordination-shape constraint, not a halt-state. Claude's Cycle 1 approval on #11670 flagged a non-blocking follow-up: the always-loaded AGENTS.md critical gate still says Self-assign + broadcast [lane-claim] to AGENT:* with no compact trigger pointer to the fallback now documented in the skill atlas.

This matters because the night-shift gridlock originated from the always-loaded map-level literal, not from the skill payloads. Agents under coordination pressure may obey rule 7 before loading /peer-role or /post-review-pickup, especially around write-lane pickup.

Duplicate sweep before filing:

• ask_knowledge_base(type:'ticket', query:'open ticket duplicate AGENTS.md critical gate rule 7 broadcast lane-claim direct DM fallback broadcast suppression halt-state') returned no relevant documents.
• GitHub issue search for critical_gates AGENT:* direct DM fallback and rule 7 lane-claim broadcast fallback AGENTS.md found #11669/#11670 plus closed lineage tickets, but no open ticket specifically targeting the rule-7 map pointer.
• Local rg over resources/content/issues, resources/content/discussions, and .agents/skills found #11345 as the closed ticket that added the current rule-7 broadcast wording, #11669 as the skill-atlas fallback ticket, and the new fallback text in peer-role / post-review-pickup. None covers this specific map-pointer decision.

The Problem

AGENTS.md:53 currently preserves the canonical collision-prevention rule:

Self-assign + broadcast [lane-claim] to AGENT:* before any git-tracked edit.

That wording is correct for normal operation, but incomplete under the operator-suppressed-broadcast incident shape. The skill-atlas fallback now says direct DM to the reachable peer is valid when the operator explicitly suppresses broadcast to protect an unstable peer harness. The always-loaded rule has no pointer to that fallback.

The risk is not that AGENT:* should stop being canonical. The risk is that the canonical rule is read as exclusive even when the operator has temporarily forbidden that channel. That recreates the false halt-state: "broadcast is forbidden, therefore work is blocked."

The Architectural Reality

Relevant current surfaces:

• AGENTS.md:53 — critical gate 7, always-loaded, keep disposition, machine-enforceable candidate.
• .agents/skills/peer-role/references/peer-role-mode.md:68 — lane-claim recipient contract now names the operator-authorized reachable peer DM fallback.
• .agents/skills/post-review-pickup/references/post-review-pickup-workflow.md:153-165 — halt-state atlas now defines the broadcast-suppressed coordination fallback.
• learn/agentos/decisions/0007-agents-md-compaction-taxonomy.md — requires any AGENTS.md change to use the 3-axis slot rule; likely disposition is compress-to-trigger, not a full rule copy.
• .agents/skills/turn-memory-pre-flight/references/turn-memory-pre-flight-workflow.md — implementation must verify placement before mutating turn-loaded substrate.

This ticket is deliberately about the map pointer only. The full fallback rule already belongs in the skill atlas and should not be duplicated into AGENTS.md.

The Fix

Evaluate and, if the 3-axis slot rule passes, add a compact trigger pointer to AGENTS.md critical gate 7. The pointer should preserve AGENT:* as canonical while making the exception discoverable.

Candidate shape, to be refined during implementation:

Self-assign + broadcast [lane-claim] to AGENT:* before any git-tracked edit; if operator-suppressed, use the documented direct-DM fallback.

Implementation must keep this a one-line compressed pointer and link the detailed rule to the skill atlas / existing enforcement chain, rather than copying the full fallback matrix into turn-loaded substrate.

Contract Ledger Matrix

Acceptance Criteria

• Run /turn-memory-pre-flight before editing AGENTS.md and cite the placement decision in the PR body.
• Load ADR 0007 and classify the rule-7 pointer as keep, compress-to-trigger, or rejected; document the 3-axis rationale.
• If accepted, update only the compact critical-gate wording needed to make the operator-suppressed fallback discoverable.
• Preserve AGENT:* as the canonical lane-claim visibility path outside explicit operator suppression.
• Do not duplicate the full broadcast-suppressed fallback matrix into AGENTS.md.
• Verify the detailed fallback remains in peer-role and post-review-pickup; update cross-references only if needed.
• Run git diff --check; run skill/agent lint if any .agents/skills/** references change.

Out of Scope

• Reopening the canonical AGENT:* lane-claim decision from #11345.
• Replacing broadcast with direct DM generally.
• Messaging Gemini while the operator's no-broadcast/no-Gemini constraint remains active.
• Runtime wake-subscription changes.
• Any code or service change.

Avoided Traps / Gold Standards Rejected

• Full-rule copy into AGENTS.md: rejected. The atlas already owns the detailed fallback; the map should only carry a compressed trigger if justified.
• Treating direct DM as the new default: rejected. The fallback is scoped to explicit operator suppression of broadcast.
• Ignoring byte budget because the follow-up is real: rejected. ADR 0007 still applies; the implementation can reject the pointer if the slot rationale fails.
• Leaving the follow-up only in PR review text: rejected. PR reviews are useful evidence, but a filed ticket is the durable work queue surface.

Related

• #11669 — skill-atlas fallback ticket
• PR #11670 — implementation and review that surfaced this follow-up
• #11345 — closed ticket that introduced the current rule-7 broadcast wording
• #10970, #11165, #11221 — post-review pickup / halt-state lineage
• ADR 0007 — compaction taxonomy for AGENTS.md mutations

Handoff Retrieval Hints

• query_raw_memories(query="AGENTS.md rule 7 broadcast fallback direct DM operator-suppressed AGENT:* lane-claim")
• query_summaries(query="nightshift watchdog broadcast suppression direct DM fallback rule 7")
• File anchors: AGENTS.md:53, .agents/skills/peer-role/references/peer-role-mode.md:68, .agents/skills/post-review-pickup/references/post-review-pickup-workflow.md:153