Post-#1174 agent-efficiency architecture execution plan

[devin-ai-integration]

Goal

Execute the remaining architecture and agent-efficiency work after #1174 without a conflict-heavy
mega-PR.

Completed foundation

• chore: add agent-efficiency regression guards #1174 — value-cycle/back-edge guards, output budgets, MCP measurement, and command ownership
  navigation.

Completed Wave 1

• Keep command owner metadata tooling-only without duplicating registries #1178 via refactor: keep command and daemon-route owner-file claims tooling-only (#1178) #1192 — owner metadata is erased from production bundles.
• Pair output budgets with workflow success and recovery benchmarks #1180 via test(output-economy): add routine-workflow output-behavior oracle #1190 — routine workflow output/actionability/recovery measurements.
• Add a derived fail-open check:affected command #1181 via feat: add derived fail-open check:affected selector #1195 — derived fail-open check:affected.
• Route the remaining perf sampling body through PlatformPlugin #1188 via refactor(daemon): route perf metrics sampling body through PlatformPlugin facet #1191 — remaining perf sampling body routed through PlatformPlugin.

Completed structural consolidation

• Eliminate remaining target-spine dependency back-edges #1182 via refactor: consolidate architecture ownership and client results #1210 — 36 target-spine back-edges reduced to zero.
• Complete public Node-client result narrowing on an explicit semver track #1183 via refactor: consolidate architecture ownership and client results #1210 — accurate closed Node-client result families narrowed; dynamic families
  remain broad by documented design.
• Split only current churn × size architecture hotspots #1184 via refactor: consolidate architecture ownership and client results #1210 and chore: close out architecture experiments #1213 — selected ownership splits completed; six-month churn × size
  remains an evidence source rather than a blanket LOC gate.
• Dissolve src/utils opportunistically into domain-owned modules #1187 via refactor: consolidate architecture ownership and client results #1210 — touched generic utilities moved to domain owners; future moves are
  opportunistic, not a mass backlog.

Completed bounded experiments

• Spike executable contract projection on one command family #1185 via Project navigation contracts and add a network digest #1208 and chore: close out architecture experiments #1213 — selected navigation declarations reduced 15 → 5; future
  projection families are independent candidates under explicit stop conditions.
• Expand digest response coverage from paired outcome evidence #1186 via Project navigation contracts and add a network digest #1208 and chore: close out architecture experiments #1213 — network digest reduced 26,371 → 1,665 bytes while preserving
  actionability and default/full compatibility; future digest families require paired
  byte/outcome evidence.

Closeout

#1213 removes the obsolete empty layering baseline and enforces the completed DAG directly. It also
records deterministic, model-runner, live-device, and churn evidence without presenting
unavailable external infrastructure as a successful experiment.

Preserved non-goals

• no command compiler, generated source, or schema DSL;
• no distributed interaction-guarantee ownership;
• no blanket LOC gate;
• no broad fast-path removal without latency/failure evidence;
• no mass move-only src/utils rewrite;
• no public ios/macos wire-value change.