| 1 | import * as NodeServices from "@effect/platform-node/NodeServices"; |
| 2 | import { assert, describe, it } from "@effect/vitest"; |
| 3 | import { |
| 4 | CommandId, |
| 5 | MessageId, |
| 6 | type ModelSelection, |
| 7 | type OrchestrationV2Command, |
| 8 | type OrchestrationV2ThreadProjection, |
| 9 | ProjectId, |
| 10 | ProviderInstanceId, |
| 11 | ProviderDriverKind, |
| 12 | type ProviderReplayTranscript, |
| 13 | ThreadId, |
| 14 | } from "@t3tools/contracts"; |
| 15 | import * as Effect from "effect/Effect"; |
| 16 | import * as FileSystem from "effect/FileSystem"; |
| 17 | import * as Layer from "effect/Layer"; |
| 18 | import * as ProviderAdapterRegistry from "../ProviderAdapterRegistry.ts"; |
| 19 | |
| 20 | import { ClaudeOrchestratorReplayHarness } from "../Adapters/ClaudeAdapterV2.testkit.ts"; |
| 21 | import { CodexOrchestratorReplayHarness } from "../Adapters/CodexAdapterV2.testkit.ts"; |
| 22 | import * as IdAllocator from "@t3tools/provider-core/server/IdAllocator"; |
| 23 | import { provideDeterministicTestRuntime } from "./DeterministicRuntime.ts"; |
| 24 | import { |
| 25 | CODEX_MODEL_SELECTION, |
| 26 | THREAD_MERGE_BACK_FORK_PROMPT, |
| 27 | THREAD_MERGE_BACK_HANDOFF_PROMPT, |
| 28 | THREAD_MERGE_BACK_RECALL, |
| 29 | THREAD_MERGE_BACK_RECALL_PROMPT, |
| 30 | THREAD_MERGE_BACK_SIBLINGS_FIRST_FORK_PROMPT, |
| 31 | THREAD_MERGE_BACK_SIBLINGS_FIRST_HANDOFF_PROMPT, |
| 32 | THREAD_MERGE_BACK_SIBLINGS_RECALL, |
| 33 | THREAD_MERGE_BACK_SIBLINGS_RECALL_PROMPT, |
| 34 | THREAD_MERGE_BACK_SIBLINGS_SECOND_FORK_PROMPT, |
| 35 | THREAD_MERGE_BACK_SIBLINGS_SECOND_HANDOFF_PROMPT, |
| 36 | THREAD_MERGE_BACK_SIBLINGS_SOURCE_PROMPT, |
| 37 | THREAD_MERGE_BACK_SOURCE_PROMPT, |
| 38 | } from "./fixtures/shared.ts"; |
| 39 | import { runOrchestratorV2ProviderReplayScenario } from "./ProviderReplayHarness.ts"; |
| 40 | import { makeCheckpointWorkspace } from "@t3tools/provider-testing/replayWorkspace"; |
| 41 | import { |
| 42 | decodeProviderReplayNdjson, |
| 43 | materializeReplayTranscriptWorkspace, |
| 44 | } from "@t3tools/provider-testing/replayTranscript"; |
| 45 | |
| 46 | // These recorded 0.137 rollouts predate injection. Preserve their native fork |
| 47 | // and turn exchanges, and assert the new history delivery at the adapter boundary. |
| 48 | // Exact thread/inject_items frames are covered by CodexAdapterV2.test.ts. |
| 49 | const CodexHistoryReplayHarness: typeof CodexOrchestratorReplayHarness = { |
| 50 | ...CodexOrchestratorReplayHarness, |
| 51 | makeProviderAdapterRegistryLayer: (transcript, options) => |
| 52 | Layer.effect( |
| 53 | ProviderAdapterRegistry.ProviderAdapterRegistryV2, |
| 54 | Effect.gen(function* () { |
| 55 | const registry = yield* ProviderAdapterRegistry.ProviderAdapterRegistryV2; |
| 56 | return { |
| 57 | ...registry, |
| 58 | get: (id) => |
| 59 | registry.get(id).pipe( |
| 60 | Effect.map((adapter) => ({ |
| 61 | ...adapter, |
| 62 | openSession: (input) => |
| 63 | adapter.openSession(input).pipe( |
| 64 | Effect.map((session) => ({ |
| 65 | ...session, |
| 66 | injectHistory: (history) => |
| 67 | Effect.sync(() => { |
| 68 | const userTexts = history.messages |
| 69 | .filter((message) => message.role === "user") |
| 70 | .map((message) => message.text); |
| 71 | assert.lengthOf(userTexts, 1); |
| 72 | assert.include( |
| 73 | [ |
| 74 | THREAD_MERGE_BACK_FORK_PROMPT, |
| 75 | THREAD_MERGE_BACK_SIBLINGS_FIRST_FORK_PROMPT, |
| 76 | THREAD_MERGE_BACK_SIBLINGS_SECOND_FORK_PROMPT, |
| 77 | ], |
| 78 | userTexts[0], |
| 79 | ); |
| 80 | assert.isTrue( |
| 81 | history.messages.some( |
| 82 | (message) => |
| 83 | message.role === "assistant" && message.text.includes("stored"), |
| 84 | ), |
| 85 | ); |
| 86 | return true; |
| 87 | }), |
| 88 | })), |
| 89 | ), |
| 90 | })), |
| 91 | ), |
| 92 | }; |
| 93 | }), |
| 94 | ).pipe( |
| 95 | Layer.provide( |
| 96 | CodexOrchestratorReplayHarness.makeProviderAdapterRegistryLayer(transcript, options), |
| 97 | ), |
| 98 | ), |
| 99 | }; |
| 100 | |
| 101 | const CLAUDE_MODEL_SELECTION = { |
| 102 | instanceId: ProviderInstanceId.make("claudeAgent"), |
| 103 | model: "claude-sonnet-4-6", |
| 104 | } as const; |
| 105 | const MERGE_BACK_USER_TEXT = |
| 106 | "Retain the transferred fork marker for later. Respond with exactly: merge delta stored"; |
| 107 | const FIRST_SIBLING_MERGE_USER_TEXT = |
| 108 | "Retain the first transferred marker for later. Respond with exactly: first merge delta stored"; |
| 109 | const SECOND_SIBLING_MERGE_USER_TEXT = |
| 110 | "Retain the second transferred marker for later. Respond with exactly: second merge delta stored"; |
| 111 | |
| 112 | interface ProviderVariant { |
| 113 | readonly driver: ProviderDriverKind; |
| 114 | readonly modelSelection: ModelSelection; |
| 115 | } |
| 116 | |
| 117 | const PROVIDERS: ReadonlyArray<ProviderVariant> = [ |
| 118 | { |
| 119 | driver: ProviderDriverKind.make("codex"), |
| 120 | // Recorded on gpt-6-sol: gpt-6-luna tends to recall an acknowledgement instead of the markers. |
| 121 | modelSelection: { ...CODEX_MODEL_SELECTION, model: "gpt-6-sol" }, |
| 122 | }, |
| 123 | { |
| 124 | driver: ProviderDriverKind.make("claudeAgent"), |
| 125 | modelSelection: CLAUDE_MODEL_SELECTION, |
| 126 | }, |
| 127 | ]; |
| 128 | |
| 129 | function transcriptPath(scenario: string, driver: ProviderDriverKind): string { |
| 130 | const fileName = driver === "codex" ? "codex_transcript.ndjson" : "claude_transcript.ndjson"; |
| 131 | return `${import.meta.dirname}/fixtures/${scenario}/${fileName}`; |
| 132 | } |
| 133 | |
| 134 | function readTranscript(scenario: string, driver: ProviderDriverKind) { |
| 135 | return Effect.gen(function* () { |
| 136 | const fs = yield* FileSystem.FileSystem; |
| 137 | const text = yield* fs.readFileString(transcriptPath(scenario, driver)); |
| 138 | return yield* decodeProviderReplayNdjson(text); |
| 139 | }); |
| 140 | } |
| 141 | |
| 142 | function compactExpectedText(text: string, maxLength = 240): string { |
| 143 | const compacted = text.replace(/\s+/gu, " ").trim(); |
| 144 | return compacted.length <= maxLength ? compacted : `${compacted.slice(0, maxLength - 3)}...`; |
| 145 | } |
| 146 | |
| 147 | function forkDeltaSummary(input: { |
| 148 | readonly sourceThreadId: ThreadId; |
| 149 | readonly targetThreadId: ThreadId; |
| 150 | readonly forkUserText: string; |
| 151 | readonly forkAssistantText: string; |
| 152 | }): string { |
| 153 | return [ |
| 154 | "Merge-back context from forked conversation.", |
| 155 | `Source thread: ${input.sourceThreadId}`, |
| 156 | `Target thread: ${input.targetThreadId}`, |
| 157 | "Covered fork runs: 1-1", |
| 158 | "", |
| 159 | "Fork delta:", |
| 160 | `- User: ${compactExpectedText(input.forkUserText)}`, |
| 161 | `- Assistant: ${compactExpectedText(input.forkAssistantText)}`, |
| 162 | "- Checkpoint: 0 files", |
| 163 | ].join("\n"); |
| 164 | } |
| 165 | |
| 166 | function providerMessage(summary: string, userText: string): string { |
| 167 | return [ |
| 168 | "Context handoff (merge_back / fork_delta_summary):", |
| 169 | summary, |
| 170 | "", |
| 171 | "User message:", |
| 172 | userText, |
| 173 | ].join("\n"); |
| 174 | } |
| 175 | |
| 176 | function replaceExactString(value: unknown, from: string, to: string): unknown { |
| 177 | if (value === from) { |
| 178 | return to; |
| 179 | } |
| 180 | if (Array.isArray(value)) { |
| 181 | return value.map((entry) => replaceExactString(entry, from, to)); |
| 182 | } |
| 183 | if (typeof value !== "object" || value === null) { |
| 184 | return value; |
| 185 | } |
| 186 | return Object.fromEntries( |
| 187 | Object.entries(value).map(([key, entry]) => [key, replaceExactString(entry, from, to)]), |
| 188 | ); |
| 189 | } |
| 190 | |
| 191 | function parameterizeHandoffs( |
| 192 | transcript: ProviderReplayTranscript, |
| 193 | replacements: ReadonlyArray<readonly [recorded: string, generated: string]>, |
| 194 | ): ProviderReplayTranscript { |
| 195 | return replacements.reduce( |
| 196 | (current, [recorded, generated]) => |
| 197 | replaceExactString(current, recorded, generated) as ProviderReplayTranscript, |
| 198 | transcript, |
| 199 | ); |
| 200 | } |
| 201 | |
| 202 | function nativeSourceThreadId(transcript: ProviderReplayTranscript): string { |
| 203 | if (transcript.provider === "claudeAgent") { |
| 204 | const value = transcript.metadata?.nativeSessionId; |
| 205 | if (typeof value !== "string") { |
| 206 | throw new Error(`${transcript.scenario} is missing metadata.nativeSessionId.`); |
| 207 | } |
| 208 | return value; |
| 209 | } |
| 210 | for (const entry of transcript.entries) { |
| 211 | if (entry.type !== "emit_inbound") { |
| 212 | continue; |
| 213 | } |
| 214 | const frame = entry.frame as { |
| 215 | readonly method?: unknown; |
| 216 | readonly params?: { readonly thread?: { readonly id?: unknown } }; |
| 217 | }; |
| 218 | if (frame.method === "thread/started" && typeof frame.params?.thread?.id === "string") { |
| 219 | return frame.params.thread.id; |
| 220 | } |
| 221 | } |
| 222 | throw new Error(`${transcript.scenario} is missing the source native thread id.`); |
| 223 | } |
| 224 | |
| 225 | function visibleConversationText(projection: OrchestrationV2ThreadProjection): string { |
| 226 | return projection.visibleTurnItems |
| 227 | .flatMap(({ item }) => |
| 228 | item.type === "user_message" || item.type === "assistant_message" ? [item.text] : [], |
| 229 | ) |
| 230 | .join("\n"); |
| 231 | } |
| 232 | |
| 233 | function normalizePipeSpacing(text: string): string { |
| 234 | return text.replace(/\s*\|\s*/gu, "|"); |
| 235 | } |
| 236 | |
| 237 | function makeCreateCommand(input: { |
| 238 | readonly commandId: CommandId; |
| 239 | readonly threadId: ThreadId; |
| 240 | readonly projectId: ProjectId; |
| 241 | readonly modelSelection: ModelSelection; |
| 242 | }): OrchestrationV2Command { |
| 243 | return { |
| 244 | type: "thread.create", |
| 245 | createdBy: "user", |
| 246 | creationSource: "web", |
| 247 | commandId: input.commandId, |
| 248 | threadId: input.threadId, |
| 249 | projectId: input.projectId, |
| 250 | title: "Merge-back source", |
| 251 | modelSelection: input.modelSelection, |
| 252 | runtimeMode: "full-access", |
| 253 | interactionMode: "default", |
| 254 | branch: null, |
| 255 | worktreePath: null, |
| 256 | }; |
| 257 | } |
| 258 | |
| 259 | describe("orchestration V2 merge-back provider replay", () => { |
| 260 | it.effect.each(PROVIDERS)( |
| 261 | "merges one fork delta back into the original $driver thread", |
| 262 | (variant) => |
| 263 | Effect.gen(function* () { |
| 264 | const rawTranscript = yield* readTranscript("thread_merge_back_continue", variant.driver); |
| 265 | const materialized = yield* Effect.gen(function* () { |
| 266 | const ids = yield* IdAllocator.IdAllocatorV2; |
| 267 | const projectId = yield* ids.allocate.project({ |
| 268 | fixtureName: `thread-merge-back-${variant.driver}`, |
| 269 | }); |
| 270 | const sourceThreadId = yield* ids.allocate.thread({ |
| 271 | fixtureName: `thread-merge-back-${variant.driver}-source`, |
| 272 | projectId, |
| 273 | }); |
| 274 | const forkThreadId = ThreadId.make(`thread-merge-back-${variant.driver}-fork`); |
| 275 | const forkRunId = ids.derive.run({ threadId: forkThreadId, ordinal: 1 }); |
| 276 | const commands = [ |
| 277 | makeCreateCommand({ |
| 278 | commandId: yield* ids.allocate.command({ |
| 279 | fixtureName: `thread-merge-back-${variant.driver}`, |
| 280 | commandName: "create", |
| 281 | }), |
| 282 | threadId: sourceThreadId, |
| 283 | projectId, |
| 284 | modelSelection: variant.modelSelection, |
| 285 | }), |
| 286 | { |
| 287 | type: "message.dispatch", |
| 288 | createdBy: "user", |
| 289 | creationSource: "web", |
| 290 | commandId: yield* ids.allocate.command({ |
| 291 | fixtureName: `thread-merge-back-${variant.driver}`, |
| 292 | commandName: "source", |
| 293 | }), |
| 294 | threadId: sourceThreadId, |
| 295 | messageId: MessageId.make(`message-merge-source-${variant.driver}`), |
| 296 | text: THREAD_MERGE_BACK_SOURCE_PROMPT, |
| 297 | attachments: [], |
| 298 | modelSelection: variant.modelSelection, |
| 299 | dispatchMode: { type: "start_immediately" }, |
| 300 | }, |
| 301 | { |
| 302 | type: "thread.fork", |
| 303 | createdBy: "user", |
| 304 | creationSource: "web", |
| 305 | commandId: CommandId.make(`command-merge-fork-${variant.driver}`), |
| 306 | sourceThreadId, |
| 307 | targetThreadId: forkThreadId, |
| 308 | sourcePoint: { type: "latest_stable" }, |
| 309 | title: "Merge-back fork", |
| 310 | }, |
| 311 | { |
| 312 | type: "message.dispatch", |
| 313 | createdBy: "user", |
| 314 | creationSource: "web", |
| 315 | commandId: yield* ids.allocate.command({ |
| 316 | fixtureName: `thread-merge-back-${variant.driver}`, |
| 317 | commandName: "fork-delta", |
| 318 | }), |
| 319 | threadId: forkThreadId, |
| 320 | messageId: MessageId.make(`message-merge-fork-${variant.driver}`), |
| 321 | text: THREAD_MERGE_BACK_FORK_PROMPT, |
| 322 | attachments: [], |
| 323 | modelSelection: variant.modelSelection, |
| 324 | dispatchMode: { type: "start_immediately" }, |
| 325 | }, |
| 326 | { |
| 327 | type: "thread.merge_back", |
| 328 | createdBy: "user", |
| 329 | creationSource: "web", |
| 330 | commandId: CommandId.make(`command-merge-back-${variant.driver}`), |
| 331 | sourceThreadId: forkThreadId, |
| 332 | targetThreadId: sourceThreadId, |
| 333 | sourcePoint: { type: "run", runId: forkRunId }, |
| 334 | }, |
| 335 | { |
| 336 | type: "message.dispatch", |
| 337 | createdBy: "user", |
| 338 | creationSource: "web", |
| 339 | commandId: yield* ids.allocate.command({ |
| 340 | fixtureName: `thread-merge-back-${variant.driver}`, |
| 341 | commandName: "consume-merge", |
| 342 | }), |
| 343 | threadId: sourceThreadId, |
| 344 | messageId: MessageId.make(`message-consume-merge-${variant.driver}`), |
| 345 | text: MERGE_BACK_USER_TEXT, |
| 346 | attachments: [], |
| 347 | modelSelection: variant.modelSelection, |
| 348 | dispatchMode: { type: "start_immediately" }, |
| 349 | }, |
| 350 | { |
| 351 | type: "message.dispatch", |
| 352 | createdBy: "user", |
| 353 | creationSource: "web", |
| 354 | commandId: yield* ids.allocate.command({ |
| 355 | fixtureName: `thread-merge-back-${variant.driver}`, |
| 356 | commandName: "recall", |
| 357 | }), |
| 358 | threadId: sourceThreadId, |
| 359 | messageId: MessageId.make(`message-merge-recall-${variant.driver}`), |
| 360 | text: THREAD_MERGE_BACK_RECALL_PROMPT, |
| 361 | attachments: [], |
| 362 | modelSelection: variant.modelSelection, |
| 363 | dispatchMode: { type: "start_immediately" }, |
| 364 | }, |
| 365 | ] satisfies ReadonlyArray<OrchestrationV2Command>; |
| 366 | return { commands, sourceThreadId, forkThreadId, forkRunId }; |
| 367 | }).pipe(Effect.provide(IdAllocator.layer), provideDeterministicTestRuntime); |
| 368 | const summary = forkDeltaSummary({ |
| 369 | sourceThreadId: materialized.forkThreadId, |
| 370 | targetThreadId: materialized.sourceThreadId, |
| 371 | forkUserText: THREAD_MERGE_BACK_FORK_PROMPT, |
| 372 | forkAssistantText: "merge fork stored", |
| 373 | }); |
| 374 | const generatedProviderMessage = providerMessage(summary, MERGE_BACK_USER_TEXT); |
| 375 | const parameterizedTranscript = parameterizeHandoffs(rawTranscript, [ |
| 376 | [ |
| 377 | THREAD_MERGE_BACK_HANDOFF_PROMPT, |
| 378 | variant.driver === "codex" ? MERGE_BACK_USER_TEXT : generatedProviderMessage, |
| 379 | ], |
| 380 | ]); |
| 381 | const cwd = yield* Effect.acquireRelease( |
| 382 | Effect.promise(() => makeCheckpointWorkspace(`merge-back-${variant.driver}`)), |
| 383 | (directory) => |
| 384 | Effect.service(FileSystem.FileSystem).pipe( |
| 385 | Effect.flatMap((fs) => fs.remove(directory, { recursive: true, force: true })), |
| 386 | Effect.orDie, |
| 387 | ), |
| 388 | ); |
| 389 | const scenario = { |
| 390 | name: `thread_merge_back_continue/${variant.driver}`, |
| 391 | commands: materialized.commands, |
| 392 | steps: [ |
| 393 | { type: "dispatch" as const, command: materialized.commands[0]!, await: true }, |
| 394 | { type: "advance_clock" as const, duration: "1 millis" as const }, |
| 395 | { type: "dispatch" as const, command: materialized.commands[1]!, await: true }, |
| 396 | { type: "await_thread_idle" as const, threadId: materialized.sourceThreadId }, |
| 397 | { type: "dispatch" as const, command: materialized.commands[2]!, await: true }, |
| 398 | { type: "dispatch" as const, command: materialized.commands[3]!, await: true }, |
| 399 | { type: "await_thread_idle" as const, threadId: materialized.forkThreadId }, |
| 400 | { type: "dispatch" as const, command: materialized.commands[4]!, await: true }, |
| 401 | { type: "dispatch" as const, command: materialized.commands[5]!, await: true }, |
| 402 | { type: "await_thread_idle" as const, threadId: materialized.sourceThreadId }, |
| 403 | { type: "dispatch" as const, command: materialized.commands[6]!, await: true }, |
| 404 | { type: "await_thread_idle" as const, threadId: materialized.sourceThreadId }, |
| 405 | ], |
| 406 | projectionThreadIds: [materialized.sourceThreadId, materialized.forkThreadId], |
| 407 | runtimePolicyOverride: { cwd }, |
| 408 | }; |
| 409 | const result = |
| 410 | variant.driver === "codex" |
| 411 | ? yield* runOrchestratorV2ProviderReplayScenario( |
| 412 | { |
| 413 | ...scenario, |
| 414 | transcript: yield* CodexOrchestratorReplayHarness.decodeTranscript( |
| 415 | materializeReplayTranscriptWorkspace(parameterizedTranscript, cwd), |
| 416 | ), |
| 417 | }, |
| 418 | CodexHistoryReplayHarness, |
| 419 | ).pipe(provideDeterministicTestRuntime) |
| 420 | : yield* runOrchestratorV2ProviderReplayScenario( |
| 421 | { |
| 422 | ...scenario, |
| 423 | transcript: |
| 424 | yield* ClaudeOrchestratorReplayHarness.decodeTranscript( |
| 425 | parameterizedTranscript, |
| 426 | ), |
| 427 | }, |
| 428 | ClaudeOrchestratorReplayHarness, |
| 429 | ).pipe(provideDeterministicTestRuntime); |
| 430 | |
| 431 | const source = result.projections.get(materialized.sourceThreadId); |
| 432 | const fork = result.projections.get(materialized.forkThreadId); |
| 433 | assert.isDefined(source); |
| 434 | assert.isDefined(fork); |
| 435 | assert.equal( |
| 436 | source.providerThreads[0]?.nativeThreadRef?.nativeId, |
| 437 | nativeSourceThreadId(rawTranscript), |
| 438 | ); |
| 439 | assert.lengthOf(source.contextHandoffs, 1); |
| 440 | assert.equal(source.contextHandoffs[0]?.summaryText, summary); |
| 441 | assert.equal(source.contextHandoffs[0]?.strategy, "fork_delta_summary"); |
| 442 | const mergeTransfer = source.contextTransfers.find( |
| 443 | (transfer) => transfer.type === "merge_back", |
| 444 | ); |
| 445 | assert.isDefined(mergeTransfer); |
| 446 | assert.equal(mergeTransfer.status, "consumed"); |
| 447 | assert.equal(mergeTransfer.sourcePoint.runId, materialized.forkRunId); |
| 448 | assert.equal(mergeTransfer.resolution?.strategy, "fork_delta_context"); |
| 449 | assert.include( |
| 450 | normalizePipeSpacing(visibleConversationText(source)), |
| 451 | THREAD_MERGE_BACK_RECALL, |
| 452 | ); |
| 453 | assert.notInclude(visibleConversationText(source), "Context handoff ("); |
| 454 | assert.include(visibleConversationText(fork), "merge fork stored"); |
| 455 | }).pipe(Effect.scoped, Effect.provide(NodeServices.layer)), |
| 456 | ); |
| 457 | |
| 458 | it.effect.each(PROVIDERS)( |
| 459 | "merges two sibling fork deltas into the original $driver thread", |
| 460 | (variant) => |
| 461 | Effect.gen(function* () { |
| 462 | const rawTranscript = yield* readTranscript("thread_merge_back_siblings", variant.driver); |
| 463 | const materialized = yield* Effect.gen(function* () { |
| 464 | const ids = yield* IdAllocator.IdAllocatorV2; |
| 465 | const fixtureName = `thread-merge-back-siblings-${variant.driver}`; |
| 466 | const projectId = yield* ids.allocate.project({ fixtureName }); |
| 467 | const sourceThreadId = yield* ids.allocate.thread({ |
| 468 | fixtureName: `${fixtureName}-source`, |
| 469 | projectId, |
| 470 | }); |
| 471 | const firstForkThreadId = ThreadId.make(`${fixtureName}-first-fork`); |
| 472 | const secondForkThreadId = ThreadId.make(`${fixtureName}-second-fork`); |
| 473 | const firstForkRunId = ids.derive.run({ threadId: firstForkThreadId, ordinal: 1 }); |
| 474 | const secondForkRunId = ids.derive.run({ threadId: secondForkThreadId, ordinal: 1 }); |
| 475 | const commandId = (commandName: string) => |
| 476 | ids.allocate.command({ fixtureName, commandName }); |
| 477 | const commands = [ |
| 478 | makeCreateCommand({ |
| 479 | commandId: yield* commandId("create"), |
| 480 | threadId: sourceThreadId, |
| 481 | projectId, |
| 482 | modelSelection: variant.modelSelection, |
| 483 | }), |
| 484 | { |
| 485 | type: "message.dispatch", |
| 486 | createdBy: "user", |
| 487 | creationSource: "web", |
| 488 | commandId: yield* commandId("source"), |
| 489 | threadId: sourceThreadId, |
| 490 | messageId: MessageId.make(`message-${fixtureName}-source`), |
| 491 | text: THREAD_MERGE_BACK_SIBLINGS_SOURCE_PROMPT, |
| 492 | attachments: [], |
| 493 | modelSelection: variant.modelSelection, |
| 494 | dispatchMode: { type: "start_immediately" }, |
| 495 | }, |
| 496 | { |
| 497 | type: "thread.fork", |
| 498 | createdBy: "user", |
| 499 | creationSource: "web", |
| 500 | commandId: yield* commandId("first-fork"), |
| 501 | sourceThreadId, |
| 502 | targetThreadId: firstForkThreadId, |
| 503 | sourcePoint: { type: "latest_stable" }, |
| 504 | title: "First merge-back sibling", |
| 505 | }, |
| 506 | { |
| 507 | type: "message.dispatch", |
| 508 | createdBy: "user", |
| 509 | creationSource: "web", |
| 510 | commandId: yield* commandId("first-fork-delta"), |
| 511 | threadId: firstForkThreadId, |
| 512 | messageId: MessageId.make(`message-${fixtureName}-first-fork`), |
| 513 | text: THREAD_MERGE_BACK_SIBLINGS_FIRST_FORK_PROMPT, |
| 514 | attachments: [], |
| 515 | modelSelection: variant.modelSelection, |
| 516 | dispatchMode: { type: "start_immediately" }, |
| 517 | }, |
| 518 | { |
| 519 | type: "thread.fork", |
| 520 | createdBy: "user", |
| 521 | creationSource: "web", |
| 522 | commandId: yield* commandId("second-fork"), |
| 523 | sourceThreadId, |
| 524 | targetThreadId: secondForkThreadId, |
| 525 | sourcePoint: { type: "latest_stable" }, |
| 526 | title: "Second merge-back sibling", |
| 527 | }, |
| 528 | { |
| 529 | type: "message.dispatch", |
| 530 | createdBy: "user", |
| 531 | creationSource: "web", |
| 532 | commandId: yield* commandId("second-fork-delta"), |
| 533 | threadId: secondForkThreadId, |
| 534 | messageId: MessageId.make(`message-${fixtureName}-second-fork`), |
| 535 | text: THREAD_MERGE_BACK_SIBLINGS_SECOND_FORK_PROMPT, |
| 536 | attachments: [], |
| 537 | modelSelection: variant.modelSelection, |
| 538 | dispatchMode: { type: "start_immediately" }, |
| 539 | }, |
| 540 | { |
| 541 | type: "thread.merge_back", |
| 542 | createdBy: "user", |
| 543 | creationSource: "web", |
| 544 | commandId: yield* commandId("merge-first"), |
| 545 | sourceThreadId: firstForkThreadId, |
| 546 | targetThreadId: sourceThreadId, |
| 547 | sourcePoint: { type: "run", runId: firstForkRunId }, |
| 548 | }, |
| 549 | { |
| 550 | type: "message.dispatch", |
| 551 | createdBy: "user", |
| 552 | creationSource: "web", |
| 553 | commandId: yield* commandId("consume-first"), |
| 554 | threadId: sourceThreadId, |
| 555 | messageId: MessageId.make(`message-${fixtureName}-consume-first`), |
| 556 | text: FIRST_SIBLING_MERGE_USER_TEXT, |
| 557 | attachments: [], |
| 558 | modelSelection: variant.modelSelection, |
| 559 | dispatchMode: { type: "start_immediately" }, |
| 560 | }, |
| 561 | { |
| 562 | type: "thread.merge_back", |
| 563 | createdBy: "user", |
| 564 | creationSource: "web", |
| 565 | commandId: yield* commandId("merge-second"), |
| 566 | sourceThreadId: secondForkThreadId, |
| 567 | targetThreadId: sourceThreadId, |
| 568 | sourcePoint: { type: "run", runId: secondForkRunId }, |
| 569 | }, |
| 570 | { |
| 571 | type: "message.dispatch", |
| 572 | createdBy: "user", |
| 573 | creationSource: "web", |
| 574 | commandId: yield* commandId("consume-second"), |
| 575 | threadId: sourceThreadId, |
| 576 | messageId: MessageId.make(`message-${fixtureName}-consume-second`), |
| 577 | text: SECOND_SIBLING_MERGE_USER_TEXT, |
| 578 | attachments: [], |
| 579 | modelSelection: variant.modelSelection, |
| 580 | dispatchMode: { type: "start_immediately" }, |
| 581 | }, |
| 582 | { |
| 583 | type: "message.dispatch", |
| 584 | createdBy: "user", |
| 585 | creationSource: "web", |
| 586 | commandId: yield* commandId("recall"), |
| 587 | threadId: sourceThreadId, |
| 588 | messageId: MessageId.make(`message-${fixtureName}-recall`), |
| 589 | text: THREAD_MERGE_BACK_SIBLINGS_RECALL_PROMPT, |
| 590 | attachments: [], |
| 591 | modelSelection: variant.modelSelection, |
| 592 | dispatchMode: { type: "start_immediately" }, |
| 593 | }, |
| 594 | ] satisfies ReadonlyArray<OrchestrationV2Command>; |
| 595 | return { |
| 596 | commands, |
| 597 | sourceThreadId, |
| 598 | firstForkThreadId, |
| 599 | secondForkThreadId, |
| 600 | firstForkRunId, |
| 601 | secondForkRunId, |
| 602 | }; |
| 603 | }).pipe(Effect.provide(IdAllocator.layer), provideDeterministicTestRuntime); |
| 604 | const firstSummary = forkDeltaSummary({ |
| 605 | sourceThreadId: materialized.firstForkThreadId, |
| 606 | targetThreadId: materialized.sourceThreadId, |
| 607 | forkUserText: THREAD_MERGE_BACK_SIBLINGS_FIRST_FORK_PROMPT, |
| 608 | forkAssistantText: "first merge sibling stored", |
| 609 | }); |
| 610 | const secondSummary = forkDeltaSummary({ |
| 611 | sourceThreadId: materialized.secondForkThreadId, |
| 612 | targetThreadId: materialized.sourceThreadId, |
| 613 | forkUserText: THREAD_MERGE_BACK_SIBLINGS_SECOND_FORK_PROMPT, |
| 614 | forkAssistantText: "second merge sibling stored", |
| 615 | }); |
| 616 | const transcript = parameterizeHandoffs(rawTranscript, [ |
| 617 | [ |
| 618 | THREAD_MERGE_BACK_SIBLINGS_FIRST_HANDOFF_PROMPT, |
| 619 | variant.driver === "codex" |
| 620 | ? FIRST_SIBLING_MERGE_USER_TEXT |
| 621 | : providerMessage(firstSummary, FIRST_SIBLING_MERGE_USER_TEXT), |
| 622 | ], |
| 623 | [ |
| 624 | THREAD_MERGE_BACK_SIBLINGS_SECOND_HANDOFF_PROMPT, |
| 625 | variant.driver === "codex" |
| 626 | ? SECOND_SIBLING_MERGE_USER_TEXT |
| 627 | : providerMessage(secondSummary, SECOND_SIBLING_MERGE_USER_TEXT), |
| 628 | ], |
| 629 | ]); |
| 630 | const cwd = yield* Effect.acquireRelease( |
| 631 | Effect.promise(() => makeCheckpointWorkspace(`merge-back-siblings-${variant.driver}`)), |
| 632 | (directory) => |
| 633 | Effect.service(FileSystem.FileSystem).pipe( |
| 634 | Effect.flatMap((fs) => fs.remove(directory, { recursive: true, force: true })), |
| 635 | Effect.orDie, |
| 636 | ), |
| 637 | ); |
| 638 | const steps = [ |
| 639 | { type: "dispatch" as const, command: materialized.commands[0]!, await: true }, |
| 640 | { type: "advance_clock" as const, duration: "1 millis" as const }, |
| 641 | { type: "dispatch" as const, command: materialized.commands[1]!, await: true }, |
| 642 | { type: "await_thread_idle" as const, threadId: materialized.sourceThreadId }, |
| 643 | { type: "dispatch" as const, command: materialized.commands[2]!, await: true }, |
| 644 | { type: "dispatch" as const, command: materialized.commands[3]!, await: true }, |
| 645 | { type: "await_thread_idle" as const, threadId: materialized.firstForkThreadId }, |
| 646 | { type: "dispatch" as const, command: materialized.commands[4]!, await: true }, |
| 647 | { type: "dispatch" as const, command: materialized.commands[5]!, await: true }, |
| 648 | { type: "await_thread_idle" as const, threadId: materialized.secondForkThreadId }, |
| 649 | { type: "dispatch" as const, command: materialized.commands[6]!, await: true }, |
| 650 | { type: "dispatch" as const, command: materialized.commands[7]!, await: true }, |
| 651 | { type: "await_thread_idle" as const, threadId: materialized.sourceThreadId }, |
| 652 | { type: "dispatch" as const, command: materialized.commands[8]!, await: true }, |
| 653 | { type: "dispatch" as const, command: materialized.commands[9]!, await: true }, |
| 654 | { type: "await_thread_idle" as const, threadId: materialized.sourceThreadId }, |
| 655 | { type: "dispatch" as const, command: materialized.commands[10]!, await: true }, |
| 656 | { type: "await_thread_idle" as const, threadId: materialized.sourceThreadId }, |
| 657 | ]; |
| 658 | const scenario = { |
| 659 | name: `thread_merge_back_siblings/${variant.driver}`, |
| 660 | commands: materialized.commands, |
| 661 | steps, |
| 662 | projectionThreadIds: [ |
| 663 | materialized.sourceThreadId, |
| 664 | materialized.firstForkThreadId, |
| 665 | materialized.secondForkThreadId, |
| 666 | ], |
| 667 | runtimePolicyOverride: { cwd }, |
| 668 | }; |
| 669 | const result = |
| 670 | variant.driver === "codex" |
| 671 | ? yield* runOrchestratorV2ProviderReplayScenario( |
| 672 | { |
| 673 | ...scenario, |
| 674 | transcript: yield* CodexOrchestratorReplayHarness.decodeTranscript( |
| 675 | materializeReplayTranscriptWorkspace(transcript, cwd), |
| 676 | ), |
| 677 | }, |
| 678 | CodexHistoryReplayHarness, |
| 679 | ).pipe(provideDeterministicTestRuntime) |
| 680 | : yield* runOrchestratorV2ProviderReplayScenario( |
| 681 | { |
| 682 | ...scenario, |
| 683 | transcript: yield* ClaudeOrchestratorReplayHarness.decodeTranscript(transcript), |
| 684 | }, |
| 685 | ClaudeOrchestratorReplayHarness, |
| 686 | ).pipe(provideDeterministicTestRuntime); |
| 687 | |
| 688 | const source = result.projections.get(materialized.sourceThreadId); |
| 689 | const firstFork = result.projections.get(materialized.firstForkThreadId); |
| 690 | const secondFork = result.projections.get(materialized.secondForkThreadId); |
| 691 | assert.isDefined(source); |
| 692 | assert.isDefined(firstFork); |
| 693 | assert.isDefined(secondFork); |
| 694 | assert.equal( |
| 695 | source.providerThreads[0]?.nativeThreadRef?.nativeId, |
| 696 | nativeSourceThreadId(rawTranscript), |
| 697 | ); |
| 698 | const firstForkNativeId = firstFork.providerThreads[0]?.nativeThreadRef?.nativeId; |
| 699 | const secondForkNativeId = secondFork.providerThreads[0]?.nativeThreadRef?.nativeId; |
| 700 | assert.isDefined(firstForkNativeId); |
| 701 | assert.isDefined(secondForkNativeId); |
| 702 | assert.notEqual(firstForkNativeId, secondForkNativeId); |
| 703 | assert.notEqual(firstForkNativeId, source.providerThreads[0]?.nativeThreadRef?.nativeId); |
| 704 | assert.notEqual(secondForkNativeId, source.providerThreads[0]?.nativeThreadRef?.nativeId); |
| 705 | assert.deepEqual( |
| 706 | source.contextHandoffs.map((handoff) => handoff.summaryText), |
| 707 | [firstSummary, secondSummary], |
| 708 | ); |
| 709 | const mergeTransfers = source.contextTransfers.filter( |
| 710 | (transfer) => transfer.type === "merge_back", |
| 711 | ); |
| 712 | assert.lengthOf(mergeTransfers, 2); |
| 713 | assert.deepEqual( |
| 714 | mergeTransfers.map((transfer) => transfer.status), |
| 715 | ["consumed", "consumed"], |
| 716 | ); |
| 717 | assert.deepEqual( |
| 718 | mergeTransfers.map((transfer) => transfer.sourcePoint.runId), |
| 719 | [materialized.firstForkRunId, materialized.secondForkRunId], |
| 720 | ); |
| 721 | assert.include( |
| 722 | normalizePipeSpacing(visibleConversationText(source)), |
| 723 | THREAD_MERGE_BACK_SIBLINGS_RECALL, |
| 724 | ); |
| 725 | assert.notInclude(visibleConversationText(source), "Context handoff ("); |
| 726 | assert.include(visibleConversationText(firstFork), "first merge sibling stored"); |
| 727 | assert.notInclude(visibleConversationText(firstFork), "second merge sibling stored"); |
| 728 | assert.include(visibleConversationText(secondFork), "second merge sibling stored"); |
| 729 | assert.notInclude(visibleConversationText(secondFork), "first merge sibling stored"); |
| 730 | }).pipe(Effect.scoped, Effect.provide(NodeServices.layer)), |
| 731 | ); |
| 732 | }); |