From 83b1a630dd6c78e479dd25097715e97b081cfb56 Mon Sep 17 00:00:00 2001 From: "lusendong.6789" Date: Sat, 14 Nov 2026 13:34:44 +0800 Subject: [PATCH] Check Todo applicability before loading research implementations Signed-off-by: lusendong.6789 Co-authored-by: TRAE CLI --- .../rfcs/monorepo-distribution-split-v0.md | 29 ++++++ .../monorepo-distribution-split-v0.zh-CN.md | 20 ++++ .../capabilities/explore_research_terminal.ts | 4 +- .../effect_runtime_handler_loading.test.ts | 96 +++++++++++++++++++ 4 files changed, 147 insertions(+), 2 deletions(-) diff --git a/docs/architecture/rfcs/monorepo-distribution-split-v0.md b/docs/architecture/rfcs/monorepo-distribution-split-v0.md index 96b3cc3a99..34d3aa06a4 100644 --- a/docs/architecture/rfcs/monorepo-distribution-split-v0.md +++ b/docs/architecture/rfcs/monorepo-distribution-split-v0.md @@ -3995,3 +3995,32 @@ owners. A valid terminal request still loads the adapter, including when its ordinary-Todo guard returns early. Complete workflow isolation, paired complete costs, real-host adoption, R1/M3 and C1 writer/consumer cutover remain open. These tests do not authorize C1 default promotion or establish model benefit. + + +### 2026-11-14 — Ordinary completion defers research implementations + +`ResearchTerminalEvidenceHost.qualify` loads its research execution and +configuration implementations after its existing Todo applicability check. +A Todo without research or replan applicability signals can therefore persist +its completion evidence and replay its original receipt while either research +implementation is unavailable. +Research-qualified Todos still select the same policy and evidence owners; +module failures retain their error and commit no completion state or receipt. +The existing source checks, writer fence, qualification, CAS and cleanup apply. + +Fresh-process fault cases cover both unavailable implementations on real File +and SQLite authority. Both retained baseline probes fail at the ordinary +completion; the candidate completes and independently reads back the result, +replays without a new effect and refuses repeated selected-research failures. +The available-module comparison passes the same 107 existing native cases; +the candidate additionally passes the failure-isolation case, for 108 total. +The real research CLI suite passes 11 cases. The same production delta and +fault cases pass with the exact C1 candidates #8266 (`b0b930bfb92a`) and #8484 +(`b7e4e01cc423`), with 190 and 189 authority, context and recovery cases. An independently built Core wheel, installed alone, also passes the same File/SQLite fault paths with production module bytes equal to the source; this does not qualify an installed model Host or adoption. + +This closes the evidenced ordinary-completion dependency failure using the +existing research guard, without duplicating the C1 work contract or its +producer/consumer pipeline. Research- or replan-associated Todos still enter the existing policy check and +load its implementations; a disabled policy does not remove that remaining +loading dependency. Complete D2/D5 closure, real-model Host/context continuity, progressive controls, +comparable complete cost, R1/M3 and C1 default cutover remain unqualified. diff --git a/docs/architecture/rfcs/monorepo-distribution-split-v0.zh-CN.md b/docs/architecture/rfcs/monorepo-distribution-split-v0.zh-CN.md index d621d18086..53173f02fa 100644 --- a/docs/architecture/rfcs/monorepo-distribution-split-v0.zh-CN.md +++ b/docs/architecture/rfcs/monorepo-distribution-split-v0.zh-CN.md @@ -3113,3 +3113,23 @@ terminal lifecycle 请求选中它时加载。普通 coordination 不再加载 适配器,包括 ordinary Todo 的 guard 随后提前返回的情况。完整工作流隔离、 成对完整成本、真实 Host 采用、R1/M3 以及 C1 writer/consumer 切换仍开放。 这些测试不授权 C1 默认推广,也不证明模型收益。 + + +### 2026-11-14 — 普通任务完成延后研究实现加载 + +`ResearchTerminalEvidenceHost.qualify` 在原有 Todo 适用性检查之后才加载研究执行与 +配置实现。因此,即使任一研究实现不可用,没有研究或重规划适用性标记的 Todo 仍能持久化 +完成证据并重放原回执。需要研究资格的 Todo 继续选中原策略和证据 owner;模块失败保留错误且不提交 +完成状态或回执。原 source 检查、writer fence、资格判定、CAS 和清理继续执行。 + +新进程故障用例在真实 File 与 SQLite authority 上覆盖两个不可用实现。两个保留的 +基线探针均在普通完成处失败;候选能够完成并独立回读结果,重放时不产生新效果,且 +重复的已选中研究失败仍被拒绝。模块可用时,同一组原有 107 项 native 用例全部通过; +候选另通过故障隔离用例,共 108 项。真实研究 CLI 套件通过 11 项。同一 production +改动与故障用例和 C1 精确候选 #8266(`b0b930bfb92a`)、#8484(`b7e4e01cc423`) +组合后,分别通过 190 与 189 项 authority、上下文和恢复用例。独立构建并单独安装的 Core wheel 也通过同一 File/SQLite 故障路径,其 production 模块字节与源码一致;这不验证模型 Host 的安装或采用。 + +这项交付通过原研究 guard 修复已证明的普通完成依赖故障,复用 C1 的工作契约与 +producer/consumer 管线。带研究或重规划标记的 Todo 仍进入原策略检查并加载其实现;策略禁用时也保留 +这项待隔离的加载依赖。完整 D2/D5 闭包、真实模型 +Host/上下文延续、渐进控制、可比完整成本、R1/M3 与 C1 默认切换仍未完成资格验证。 diff --git a/loopx/control_plane/capabilities/explore_research_terminal.ts b/loopx/control_plane/capabilities/explore_research_terminal.ts index 3015635626..7d583f0293 100644 --- a/loopx/control_plane/capabilities/explore_research_terminal.ts +++ b/loopx/control_plane/capabilities/explore_research_terminal.ts @@ -9,8 +9,6 @@ import {createInterface} from "node:readline"; import type {JsonObject} from "../effect_program.ts"; import type {TodoTerminalEvidenceGuard} from "../coordination/todo_terminal_lifecycle.ts"; import {requireJsonObject} from "../runtime_decode.ts"; -import {normalizeResearchCompositionPolicy, qualifyResearchCompletion} from "./explore_research_execution.ts"; -import {resolveExploreGoalConfiguration} from "./explore_configuration.ts"; export class ResearchTerminalEvidenceHost { private child: ChildProcessWithoutNullStreams | null = null; @@ -26,6 +24,8 @@ export class ResearchTerminalEvidenceHost { && context.todo.capability_binding_ref !== "explore:research-composition-v0" && !(Array.isArray(context.todo.explore_result_node_refs) && context.todo.explore_result_node_refs.length) && !context.todo.replan_obligation_id) return {required: false, allowed: true, evidence: null}; + const {normalizeResearchCompositionPolicy, qualifyResearchCompletion} = await import("./explore_research_execution.ts"); + const {resolveExploreGoalConfiguration} = await import("./explore_configuration.ts"); const source = requireJsonObject(this.source, "research registry source"); const bytes = await readFile(String(source.path)); if (createHash("sha256").update(bytes).digest("hex") !== source.sha256) { diff --git a/tests/control_plane_ts/effect_runtime_handler_loading.test.ts b/tests/control_plane_ts/effect_runtime_handler_loading.test.ts index 1b435abb31..31bb709f48 100644 --- a/tests/control_plane_ts/effect_runtime_handler_loading.test.ts +++ b/tests/control_plane_ts/effect_runtime_handler_loading.test.ts @@ -88,6 +88,102 @@ test("coordination claims remain available without terminal research evidence", assert.equal(child.status, 0, child.stderr); }); +test("ordinary completion stays available without research implementation modules", () => { + const entry = new URL("../../loopx/control_plane/effect_runtime_handlers.ts", import.meta.url).href; + const fixtures = new URL("./local_promotion_fixture.ts", import.meta.url).href; + const source = new URL("./shadow_file_fixture.ts", import.meta.url).href; + const storeModule = new URL("../../loopx/control_plane/coordination/file_authority_store.ts", import.meta.url).href; + const providerModule = new URL("../../loopx/control_plane/coordination/local_authority_provider.ts", import.meta.url).href; + for (const backend of ["file", "sqlite"]) { + for (const blockedModule of ["explore_research_execution.ts", "explore_configuration.ts"]) { + const script = ` + import assert from "node:assert/strict"; + import {registerHooks} from "node:module"; + import {mkdtemp, rm, writeFile} from "node:fs/promises"; + import {createHash} from "node:crypto"; + import {tmpdir} from "node:os"; + import {join} from "node:path"; + let blockedLoads = 0; + registerHooks({load(url, context, nextLoad) { + if (url.endsWith("/capabilities/" + ${JSON.stringify(blockedModule)})) { + blockedLoads++; + throw new Error("research implementation unavailable"); + } + return nextLoad(url, context); + }}); + const {createEffectRuntimeHandlers, dispatchEffectRuntimeMethod} = await import(${JSON.stringify(entry)}); + const {engageCanonicalRoute} = await import(${JSON.stringify(fixtures)}); + const {projection, todo} = await import(${JSON.stringify(source)}); + const {FileAuthorityStore} = await import(${JSON.stringify(storeModule)}); + const root = await mkdtemp(join(tmpdir(), "coordination-ordinary-completion-")); + try { + let store; + if (${JSON.stringify(backend)} === "sqlite") { + const {selectLocalSqliteAuthority, openLocalAuthorityStore} = await import(${JSON.stringify(providerModule)}); + await selectLocalSqliteAuthority(root, "goal-a", true); + store = await openLocalAuthorityStore(root, "goal-a"); + } else { + store = new FileAuthorityStore(join(root, "authority", "file-v0"), "goal-a"); + } + await engageCanonicalRoute(root); + const ordinary = {...todo(), claimed_by: "agent-a"}; + const research = {...todo("todo_research"), claimed_by: "agent-a", action_kind: "joint_probe"}; + assert.equal((await store.commitAuthority({expected_provider_revision: null, + operation_id: "seed", events: [], receipts: [], + next_projection: projection([ordinary, research], [], "soft_claim")})).status, "applied"); + const registry = join(root, "registry.json"); + await writeFile(registry, "{}"); + const handlers = createEffectRuntimeHandlers({fingerprint: "loading-test", requestShutdown() {}}); + const request = { + schema_version: "loopx_local_coordination_todo_terminal_lifecycle_request_v3", + registry_source: {path: registry, sha256: createHash("sha256").update("{}").digest("hex")}, + runtime_root: root, goal_id: "goal-a", todo_id: "todo_one", role: "agent", + command: "complete", actor_agent_id: "agent-a", registered_agents: ["agent-a"], + lifecycle_grants: [], authority_reason: null, decision_outcome: null, + operation_identity: {kind: "explicit", operation_id: "ordinary-complete"}, + lease_idempotency_key: null, lease_expected_version: null, allow_user_gate_auto_acquire: false, + requested_no_followup: true, requested_completion_turn_key: null, + requested_completion_identity_source: null, linked_successor_todo_ids: [], successor_intents: [], + note: null, evidence: "validated ordinary result", reason: null, clear_claim: false, + validation_declaration: null, validation_receipt: null, completion_policy_request: null, + observed_at: "2026-09-05T04:30:00Z", dry_run: false, + }; + const completed = await dispatchEffectRuntimeMethod(handlers, + "coordination.local_authority.todo_terminal", request); + assert.equal(completed.status, "applied", JSON.stringify(completed)); + const accepted = await store.loadAuthority(); + assert.equal(accepted.status, "loaded"); + const result = accepted.head.todos.find(row => row.todo_id === "todo_one"); + assert.equal(result.status, "done"); + assert.equal(result.done, true); + assert.equal(result.evidence, "validated ordinary result"); + const receipt = await store.readReceipt("ordinary-complete"); + assert.equal(receipt.status, "found"); + assert.equal((await dispatchEffectRuntimeMethod(handlers, + "coordination.local_authority.todo_terminal", request)).status, "replayed"); + assert.deepEqual(await store.loadAuthority(), accepted); + assert.deepEqual(await store.readReceipt("ordinary-complete"), receipt); + assert.equal(blockedLoads, 0); + const selected = {...request, todo_id: "todo_research", + operation_identity: {kind: "explicit", operation_id: "research-complete"}}; + for (let attempt = 0; attempt < 2; attempt++) { + const failed = await dispatchEffectRuntimeMethod(handlers, + "coordination.local_authority.todo_terminal", selected); + assert.equal(failed.status, "failed"); + assert.match(failed.reason, /research implementation unavailable/); + assert.deepEqual(await store.readReceipt("research-complete"), {status: "missing"}); + assert.deepEqual(await store.loadAuthority(), accepted); + } + assert.equal((await dispatchEffectRuntimeMethod(handlers, "runtime.ping", {})).ready, true); + } finally {await rm(root, {recursive: true, force: true});} + `; + const child = spawnSync(process.execPath, ["--no-warnings", "--experimental-sqlite", "--experimental-strip-types", + "--input-type=module", "-e", script], {encoding: "utf8", timeout: 30_000}); + assert.equal(child.status, 0, `${backend}/${blockedModule}: ${child.stderr}`); + } + } +}); + test("runtime loads selected owners without importing unrelated methods", () => { const entry = new URL("../../loopx/control_plane/effect_runtime_handlers.ts", import.meta.url).href; // Observe a fresh process, not this test runner's already imported modules.