Sub 5: Layer 4 — structured wake metadata schema extension

Premise

Per Epic AC6: extend wake-message schema with structured metadata fields so the receiving agent can program-against the wake context instead of parsing free-form text:

{
  type                    : 'idle-out-nudge' | 'sunset-handover' | 'review-request' | ...,
  expectedAction          : '<machine-readable next-step>',
  currentLifecycleState   : '<agent's known lifecycle state from sender's POV>',
  suggestedQueries        : ['gh issue list --search ...', ...],
  currentSubstrateSnapshot: {openPrCount, unassignedCurrentReleaseCount, ...}
}

This is higher-lift than Subs 1-4 (schema change touching A2A wire format + every wake-sender + every wake-receiver).

Prescription

Schema extension to A2A wake-message format. Update:

• add_message MCP tool to accept structured metadata fields
• Wake-sender call sites to populate metadata
• Receiver-side parsing/dispatch on structured fields

Acceptance Criteria (when reactivated)

• AC1: Wake-message schema extended per template above
• AC2: Backward-compatible (existing free-form-text wakes still flow through unchanged)
• AC3: At least 2 receiver implementations consume structured fields (proves the contract is loadbearing not over-engineering)

Avoided Traps

• ❌ Implementing now without empirical evidence that Layers 1-3 + 5 are insufficient (Plan-agent over-engineering)
• ❌ Wake-schema-as-future-proof-substrate without consumer-side validation (substrate without consumer = substrate bloat per feedback_truth_in_code.md)

Depends on

• Subs 1-4 (Layers 1-3 + 5) merged + empirical observation of residual idle-out

Related

• Epic #11829 (parent, AC6 — explicitly deferred)
• Sub 1 #11905 (Layer 2 — must land first)
• Sub 2 #11906 (Layer 1 wake-content — likely covers most of the structured-metadata use cases via content-rich prompt)

Authored by: Claude Opus 4.7 (Claude Code)