fix(core): align Moirai F1 judgment contracts and Codex roadmap

AGENTS.md

@@ -1,10 +1,10 @@
 # Lina — agent guide
 
 - Read [POLICY.md](POLICY.md) for development, CI, merge and release authority; [CONTRIBUTING.md](CONTRIBUTING.md) is the contribution entry point. Keep repository rules here, not in global client settings.
-- Engine work follows the [Moirai engine design](docs/MOIRAI_ENGINE.md): three judgment modules with distinct objectives (Clotho: future outcomes, Lachesis: desire and learned preference, Atropos: adopted intentions), Moirai arbitration under a declared policy, Host ownership of records and handoff, and existing owners keeping their sources. The design is confirmed but not implemented; it is not evidence of orchestration, circuit learning or qualification.
+- Engine work follows the [Moirai engine design](docs/MOIRAI_ENGINE.md): three judgment modules with distinct objectives (Clotho: future outcomes, Lachesis: desire and learned preference, Atropos: adopted intentions), Moirai arbitration under a declared policy, Host ownership of records and handoff, and existing owners keeping their sources. F1 common contracts are implemented; product integration remains planned, and this is not evidence of orchestration, circuit learning or qualification.
 - Bun workspace; `bun test` / `bun run typecheck` / `bun run lint` from the root.
 - Strict TS (see tsconfig.json), Biome, no default exports.
-- Current runtime execution is Codex: `lina-codex` owns its RPC adapter and OpenCodex owns provider routing. The confirmed engine design (D13, D21) selects Senpi as the cognition backend, `omo app-server` as the development-task execution backend behind the existing `TaskManager` contract, and Senpi native provider accounts instead of OpenCodex; Codex CLI and OpenCodex are retired only when the F1–F4 roadmap lands that switch, not through ad-hoc edits. `lina-runtime/src/host.ts` contains Lina-owned contracts; do not introduce an engine SDK dependency into channels or memory.
+- Codex owns current and planned cognition/task execution: `lina-codex` owns the RPC adapter and TaskManager contract; OpenCodex owns provider routing. Moirai D22 supersedes the Senpi/OMO migration in D13/D21. Preserve Codex/OpenCodex and existing data. `lina-runtime/src/host.ts` contains Lina-owned contracts; do not introduce an engine SDK dependency into channels or memory.
 - Discord layering: `discord.js` only in `discord/gateway-source.ts`; `lina-channels` never imports `lina-runtime`. QA scripts under `scripts/qa/` are exempt from the discord.js restriction.
 - Every behavior change starts with a failing test; tests await signals, never sleep.
 - Runtime artifacts (`.lina-sessions`, `data/memory`, `data/notepad.md`) are gitignored.

docs/ARCHITECTURE.md

@@ -2,7 +2,7 @@
 
 Lina owns persistent agent conversations, persona and memory, task coordination and the user interface. Codex is the execution engine; OpenCodex owns provider routing. The current execution and configuration contract is [CODEX_RUNTIME](CODEX_RUNTIME.md). [Product plans](PLANNING.md) distinguish current behavior from proposed UI, desktop, history and world features.
 
-The agent core is the [Moirai engine](MOIRAI_ENGINE.md): three judgment modules with distinct objectives (Clotho: future outcomes, Lachesis: desire and learned preference via a MaleCNS mushroom-body subcircuit, Atropos: adopted intentions and continuity), Moirai arbitration under a declared per-catalog policy, and a Host that validates, records and hands off to existing owners. Its selected cognition backend is Senpi; development tasks move to `omo app-server` behind the existing TaskManager contract, and provider accounts, model listing and usage move to Senpi native (`account/*`, `model/list`), retiring Codex CLI and OpenCodex once the F1–F4 roadmap lands that switch (D21). [030](plans/context-engines/030_moirai_refactor_plan.md) keeps engine activity behind the ordinary conversation and replaces user-composed backend/model settings with maintainer-qualified presets. The [R0 QA probe](plans/context-engines/031_moirai_r0_evidence.md) exercises four persistent native threads separately from the product. Engine implementation, preset enforcement and the new conversation boundary remain planned; the Codex execution and model APIs described below still apply to the current runtime.
+The agent core is the [Moirai engine](MOIRAI_ENGINE.md): three judgment modules with distinct objectives (Clotho: future outcomes, Lachesis: desire and learned preference via a MaleCNS mushroom-body subcircuit, Atropos: adopted intentions and continuity), Moirai arbitration under a declared per-catalog policy, and a Host that validates, records and hands off to existing owners. Its current and selected cognition/task backend is Codex, with the existing lina-codex RPC adapter and TaskManager. OpenCodex retains provider accounts, routing, model listing and usage ownership. D22 supersedes the Senpi/OMO migration in D13/D21. F1 judgment contracts are implemented; D23 adds bounded prose preparation and policy revision 2 while retaining historical revision-1 replay. [030](plans/context-engines/030_moirai_refactor_plan.md) keeps engine activity behind the ordinary conversation and replaces user-composed backend/model settings with maintainer-qualified presets. The [R0 QA probe](plans/context-engines/031_moirai_r0_evidence.md) exercises four persistent native threads separately from the product. Product mechanism integration, preset enforcement and the new conversation boundary remain planned; the Codex execution and model APIs described below still apply to the current runtime.
 
 ```text
 browser -> same-origin gateway -> Lina agent/task controller