diff --git a/apps/daemon/tests/chat-route.test.ts b/apps/daemon/tests/chat-route.test.ts index 5d627d8f3..2df4b60e0 100644 --- a/apps/daemon/tests/chat-route.test.ts +++ b/apps/daemon/tests/chat-route.test.ts @@ -1,4 +1,5 @@ import type http from 'node:http'; +import Database from 'better-sqlite3'; import { randomUUID } from 'node:crypto'; import { chmodSync, @@ -27,6 +28,7 @@ import { import { skillCwdAliasSegment } from '../src/cwd-aliases.js'; import { getAgentDef } from '../src/agents.js'; import { readMemoryConfig, writeMemoryConfig } from '../src/memory.js'; +import { upsertMessage } from '../src/db.js'; import { renderCodexImagegenOverride } from '../src/prompts/system.js'; const FAKE_VELA_FIXTURE = resolve(process.cwd(), 'tests', 'fixtures', 'fake-vela.mjs'); @@ -216,6 +218,87 @@ process.exit(0); ); }); + + it('reuses an existing assistant message row instead of creating a duplicate when assistantMessageId is supplied', async () => { + if (!process.env.OD_DATA_DIR) { + throw new Error('OD_DATA_DIR is required for assistant message reuse tests'); + } + const projectId = `proj-${randomUUID()}`; + const assistantMessageId = `assistant-${randomUUID()}`; + + const createProjectResponse = await fetch(`${baseUrl}/api/projects`, { + method: 'POST', + headers: { 'Content-Type': 'application/json' }, + body: JSON.stringify({ id: projectId, name: 'Assistant row reuse fixture' }), + }); + expect(createProjectResponse.ok).toBe(true); + + const conversationsResponse = await fetch(`${baseUrl}/api/projects/${projectId}/conversations`); + expect(conversationsResponse.ok).toBe(true); + const conversationsBody = await conversationsResponse.json() as { + conversations: Array<{ id: string }>; + }; + const conversationId = conversationsBody.conversations[0]?.id; + expect(conversationId).toBeTruthy(); + + const dbFile = resolve(process.env.OD_DATA_DIR, 'app.sqlite'); + const sqlite = new Database(dbFile); + try { + upsertMessage(sqlite as never, conversationId!, { + id: assistantMessageId, + role: 'assistant', + content: '', + runStatus: 'failed', + startedAt: Date.now() - 1_000, + endedAt: Date.now() - 500, + }); + } finally { + sqlite.close(); + } + + await withFakeAgent( + 'opencode', + ` +process.stdin.resume(); +process.stdin.on('end', () => { + console.log(JSON.stringify({ type: 'step_start' })); + console.log(JSON.stringify({ type: 'text', part: { text: 'reused-assistant-row-ok' } })); + console.log(JSON.stringify({ type: 'step_finish', part: { tokens: { input: 1, output: 1 } } })); + process.exit(0); +}); +`, + async () => { + const response = await fetch(`${baseUrl}/api/chat`, { + method: 'POST', + headers: { 'Content-Type': 'application/json' }, + body: JSON.stringify({ + agentId: 'opencode', + projectId, + conversationId, + assistantMessageId, + message: 'retry this turn', + }), + }); + const body = await response.text(); + expect(response.ok).toBe(true); + expect(body).toContain('reused-assistant-row-ok'); + }, + ); + + const verifyDb = new Database(dbFile, { readonly: true }); + try { + const rows = verifyDb + .prepare(`SELECT id, content, run_id FROM messages WHERE conversation_id = ? AND role = 'assistant'`) + .all(conversationId) as Array<{ id: string; content: string; run_id: string | null }>; + expect(rows.filter((row) => row.id === assistantMessageId)).toHaveLength(1); + expect(rows.some((row) => row.id !== assistantMessageId && row.content.includes('reused-assistant-row-ok'))).toBe(false); + const reused = rows.find((row) => row.id === assistantMessageId); + expect(reused?.content).toContain('reused-assistant-row-ok'); + } finally { + verifyDb.close(); + } + }); + it('rewrites the OpenCode scanner overflow into a generic retry message', async () => { const conversationId = `conv-${randomUUID()}`; diff --git a/apps/daemon/tests/integrations/vela.routes.test.ts b/apps/daemon/tests/integrations/vela.routes.test.ts index ded918bda..a239a343c 100644 --- a/apps/daemon/tests/integrations/vela.routes.test.ts +++ b/apps/daemon/tests/integrations/vela.routes.test.ts @@ -371,6 +371,61 @@ describe('POST /api/integrations/vela/login', () => { } }); + + it('uses the same Settings-configured AMR env for login and subsequent status reads', async () => { + const dataDir = process.env.OD_DATA_DIR as string; + const previous = await readAppConfig(dataDir); + process.env.OPEN_DESIGN_AMR_PROFILE = 'prod'; + process.env.VELA_PROFILE = 'prod'; + process.env.FAKE_VELA_LOGIN_USER_EMAIL = 'settings-roundtrip@example.com'; + await writeAppConfig(dataDir, { + ...previous, + agentCliEnv: { + ...(previous.agentCliEnv ?? {}), + amr: { + ...((previous.agentCliEnv?.amr as Record) ?? {}), + VELA_BIN: FAKE_VELA, + OPEN_DESIGN_AMR_PROFILE: 'local', + }, + }, + }); + try { + const before = await getJson<{ + loggedIn: boolean; + profile: string; + user: { email?: string } | null; + }>(`${baseUrl}/api/integrations/vela/status`); + expect(before.status).toBe(200); + expect(before.body.loggedIn).toBe(false); + expect(before.body.profile).toBe('local'); + + const login = await postJson<{ + pid: number; + profile: string; + }>(`${baseUrl}/api/integrations/vela/login`); + expect(login.status).toBe(202); + expect(login.body.profile).toBe('local'); + + for (let i = 0; i < 50; i += 1) { + const current = await getJson<{ + loggedIn: boolean; + profile: string; + user: { email?: string } | null; + }>(`${baseUrl}/api/integrations/vela/status`); + if (current.body.loggedIn) { + expect(current.body.profile).toBe('local'); + expect(current.body.user?.email).toBe('settings-roundtrip@example.com'); + return; + } + await new Promise((resolve) => setTimeout(resolve, 100)); + } + throw new Error('expected configured-profile AMR login to become visible via /status'); + } finally { + await writeAppConfig(dataDir, previous as unknown as Record); + delete process.env.FAKE_VELA_LOGIN_USER_EMAIL; + } + }); + it('returns 409 when a login subprocess is already in flight', async () => { // Use the stub's delay knob so the first login is still running when // the second request arrives; without this the first exits before the diff --git a/apps/daemon/tests/runs.test.ts b/apps/daemon/tests/runs.test.ts index 9855a7de7..91651d1a3 100644 --- a/apps/daemon/tests/runs.test.ts +++ b/apps/daemon/tests/runs.test.ts @@ -62,6 +62,27 @@ describe('chat run service shutdown', () => { runs.list({ projectId: 'project-1', conversationId: 'conv-b', status: 'active' }), ).toEqual([runB]); }); + it('cancels a queued run immediately without waiting for child process shutdown', async () => { + const runs = createRuns(); + const run = runs.create({ projectId: 'project-1', conversationId: 'conv-queued' }); + + const wait = runs.wait(run); + runs.cancel(run); + + expect(run.status).toBe('canceled'); + expect(run.cancelRequested).toBe(true); + expect(run.signal).toBe('SIGTERM'); + expect(run.events.at(-1)).toMatchObject({ + event: 'end', + data: { status: 'canceled', signal: 'SIGTERM' }, + }); + await expect(wait).resolves.toMatchObject({ + status: 'canceled', + signal: 'SIGTERM', + }); + }); + + it('stores effective media execution policy on run status bodies', () => { const runs = createRuns(); diff --git a/apps/daemon/tests/runtimes/agent-args.test.ts b/apps/daemon/tests/runtimes/agent-args.test.ts index 3021632e7..c7d12d2d8 100644 --- a/apps/daemon/tests/runtimes/agent-args.test.ts +++ b/apps/daemon/tests/runtimes/agent-args.test.ts @@ -1,7 +1,7 @@ import { existsSync, readFileSync } from 'node:fs'; import { test } from 'vitest'; import { - AGENT_DEFS, aider, antigravity, assert, claude, codex, copilot, cursorAgent, deepseek, devin, detectAgents, gemini, join, kilo, kiro, mkdtempSync, opencode, pi, qoder, qwen, rmSync, spawnEnvForAgent, tmpdir, vibe, writeFileSync, chmodSync, + AGENT_DEFS, aider, antigravity, assert, claude, codex, copilot, cursorAgent, deepseek, devin, detectAgents, gemini, grokBuild, join, kilo, kiro, mkdtempSync, opencode, pi, qoder, qwen, rmSync, spawnEnvForAgent, tmpdir, vibe, writeFileSync, chmodSync, } from './helpers/test-helpers.js'; import { writeAntigravityModelSelection } from '../../src/runtimes/defs/antigravity.js'; import type { TestAgentDef } from './helpers/test-helpers.js'; @@ -756,6 +756,29 @@ test('codex buildArgs omits model_reasoning_effort when reasoning is "default"', ); }); +test('grok-build inlines the prompt as -p and never falls back to stdin sentinels', () => { + const prompt = 'summarize the current page layout'; + const args = grokBuild.buildArgs( + prompt, + [], + [], + { model: 'grok-4.3', reasoning: 'high' }, + { cwd: '/tmp/od-project' }, + ); + + assert.equal(grokBuild.promptViaStdin, false); + assert.deepEqual(args, [ + '-p', + prompt, + '--model', + 'grok-4.3', + '--effort', + 'high', + ]); + assert.equal(args.includes('-'), false); + assert.equal(args.filter((entry) => entry === '-p').length, 1); +}); + test('claude flags promptViaStdin and never embeds the prompt in argv', () => { // Long composed prompts (system prompt + design system + skill body + // user message) routinely exceed Linux MAX_ARG_STRLEN (~128 KB) and the diff --git a/apps/daemon/tests/runtimes/env-and-detection.test.ts b/apps/daemon/tests/runtimes/env-and-detection.test.ts index 82cbf78fd..0cece5a16 100644 --- a/apps/daemon/tests/runtimes/env-and-detection.test.ts +++ b/apps/daemon/tests/runtimes/env-and-detection.test.ts @@ -478,6 +478,56 @@ fsTest('detectAgents marks AMR available from packaged built-in Vela with the bu } }); + +fsTest('detectAgents prefers configured AMR live models over stale fallback defaults', async () => { + const root = mkdtempSync(join(tmpdir(), 'od-detect-amr-live-models-')); + try { + return await withEnvSnapshot(['PATH', 'OD_AGENT_HOME', 'OD_RESOURCE_ROOT', 'VELA_OPENCODE_BIN'], async () => { + const fakeVela = join(root, 'vela'); + const fakeOpenCode = join(root, 'opencode'); + writeFileSync( + fakeVela, + `#!/bin/sh +if [ "$1" = "--version" ]; then echo "vela custom-live"; exit 0; fi +if [ "$1" = "models" ]; then printf "%s\n" "public_model_deepseek_v4_flash vela" "public_model_glm_5 vela"; exit 0; fi +exit 0 +`, + ); + writeFileSync(fakeOpenCode, `#!/bin/sh +exit 0 +`); + chmodSync(fakeVela, 0o755); + chmodSync(fakeOpenCode, 0o755); + process.env.PATH = ''; + process.env.OD_AGENT_HOME = join(root, 'empty-home'); + delete process.env.OD_RESOURCE_ROOT; + delete process.env.VELA_OPENCODE_BIN; + + const agents = await detectAgents({ + amr: { + VELA_BIN: fakeVela, + VELA_OPENCODE_BIN: fakeOpenCode, + }, + }); + const amrAgent = agents.find((agent) => agent.id === 'amr'); + + assert.ok(amrAgent); + assert.equal(amrAgent.available, true); + assert.equal(amrAgent.path, fakeVela); + assert.equal(amrAgent.version, 'vela custom-live'); + assert.equal(amrAgent.modelsSource, 'live'); + assert.deepEqual(amrAgent.models, [ + { id: 'deepseek-v4-flash', label: 'deepseek-v4-flash' }, + { id: 'glm-5', label: 'glm-5' }, + ]); + assert.equal(amrAgent.models.some((model) => model.id === 'default'), false); + assert.equal(amrAgent.models.some((model) => model.id === 'gpt-5.4-mini'), false); + }); + } finally { + rmSync(root, { recursive: true, force: true }); + } +}); + function codexNativeTargetTriple(): string { if (process.platform === 'darwin' && process.arch === 'arm64') return 'aarch64-apple-darwin'; if (process.platform === 'darwin' && process.arch === 'x64') return 'x86_64-apple-darwin'; diff --git a/docs/testing/launch-review-e2e-regressions.zh-CN.md b/docs/testing/launch-review-e2e-regressions.zh-CN.md index eaa938fbf..279c28531 100644 --- a/docs/testing/launch-review-e2e-regressions.zh-CN.md +++ b/docs/testing/launch-review-e2e-regressions.zh-CN.md @@ -320,7 +320,7 @@ pnpm exec playwright test -c playwright.config.ts ui/design-systems-manager.test ## 新增 daemon 契约回归 -这批 launch review 补测不只停留在 Playwright。对于前端 E2E 无法替代的契约层问题，当前已补 5 条 daemon 定向回归。 +这批 launch review 补测不只停留在 Playwright。对于前端 E2E 无法替代的契约层问题，当前已补 10 条 daemon 定向回归。 ### 1. Diagnostics 导出路径与缺失日志清单 @@ -392,6 +392,82 @@ pnpm exec playwright test -c playwright.config.ts ui/design-systems-manager.test - `generated-plugin/SKILL.md` - daemon 会把本轮转成 `failed`，而不是错误地保留 `succeeded` + +### 6. Grok Build prompt inline argv 契约 + +文件： +- [apps/daemon/tests/runtimes/agent-args.test.ts](/Users/mac/open-design/open-design-amr-runtime-acp/apps/daemon/tests/runtimes/agent-args.test.ts) +- [apps/daemon/src/runtimes/defs/grok-build.ts](/Users/mac/open-design/open-design-amr-runtime-acp/apps/daemon/src/runtimes/defs/grok-build.ts) + +新增覆盖： + +1. `grok-build inlines the prompt as -p and never falls back to stdin sentinels` + - 覆盖近期 `grok-build` 适配器从 stdin 路径切到 `-p ` 的真实契约 + - 明确要求： + - `promptViaStdin = false` + - argv 中必须带 `-p` 与 prompt 正文 + - 不能出现旧的 `-` stdin 哨兵 + - 防止后续 refactor 把 Grok Build 又错误退回到 stdin / 占位符路径 + +### 7. AMR login/status 同源配置契约 + +文件： +- [apps/daemon/tests/integrations/vela.routes.test.ts](/Users/mac/open-design/open-design-amr-runtime-acp/apps/daemon/tests/integrations/vela.routes.test.ts) + +新增覆盖： + +1. `uses the same Settings-configured AMR env for login and subsequent status reads` + - 覆盖 `/api/integrations/vela/login` 与 `/api/integrations/vela/status` 都必须使用同一份 `agentCliEnv.amr` + - 避免终端里的 `vela` 已登录，但 Open Design 因配置源不一致读成未登录 + +### 8. queued run 取消的终态语义 + +文件： +- [apps/daemon/tests/runs.test.ts](/Users/mac/open-design/open-design-amr-runtime-acp/apps/daemon/tests/runs.test.ts) + +新增覆盖： + +1. `cancels a queued run immediately without waiting for child process shutdown` + - 覆盖 queued run 在尚未启动子进程时被取消的路径 + - 要求 run 立即进入 `canceled` 终态、发出 `end` 事件并解析 waiter，而不是悬挂在 active 状态 + +### 9. /api/agents 的 AMR live model discovery 不得回退假默认 + +文件： +- [apps/daemon/tests/runtimes/env-and-detection.test.ts](/Users/mac/open-design/open-design-amr-runtime-acp/apps/daemon/tests/runtimes/env-and-detection.test.ts) + +新增覆盖： + +1. `detectAgents prefers configured AMR live models over stale fallback defaults` + - 覆盖 `detectAgents({ amr: { VELA_BIN, VELA_OPENCODE_BIN } })` 走真实 live catalog + - 要求 `modelsSource = live` 且返回归一化后的可运行模型 id + - 明确禁止回退到 `default` / 旧的 `gpt-5.4-mini` 伪默认模型 + +### 10. retry 复用已有 assistant message 行而不是复制新行 + +文件： +- [apps/daemon/tests/chat-route.test.ts](/Users/mac/open-design/open-design-amr-runtime-acp/apps/daemon/tests/chat-route.test.ts) + +新增覆盖： + +1. `reuses an existing assistant message row instead of creating a duplicate when assistantMessageId is supplied` + - 覆盖 retry / 续跑场景下显式传入 `assistantMessageId` 时，daemon 会复用原有 assistant message 行 + - 要求同一 `conversation_id` 下该 assistant message 仍然只有一条记录 + - 新输出会回填到原行，而不是额外插入一条新的 assistant message + +### 已覆盖的 artifact quiet-period / watchdog 收尾 + +文件： +- [apps/daemon/tests/chat-run-artifact-quiet-period.test.ts](/Users/mac/open-design/open-design-amr-runtime-acp/apps/daemon/tests/chat-run-artifact-quiet-period.test.ts) + +现有覆盖已经锁住： + +- `returns succeeded when the watchdog-initiated quiet-period SIGTERM fires` +- `returns succeeded when the watchdog quiet-period escalates to SIGKILL` +- `returns failed when SIGTERM/SIGKILL arrive but no quiet-period shutdown was requested` + +这部分本轮没有重复补新测试，避免和现有专用回归重叠。 + ### daemon 定向运行命令 仓库根目录： @@ -401,7 +477,19 @@ cd /Users/mac/open-design/open-design-amr-runtime-acp/apps/daemon ``` ```bash -pnpm exec vitest run tests/chat-route.test.ts tests/diagnostics-export.test.ts tests/projects-routes.test.ts tests/runs.test.ts tests/amr-acp-integration.test.ts tests/runtimes/env-and-detection.test.ts tests/runtimes/resolve-model.test.ts +pnpm exec vitest run tests/chat-route.test.ts tests/diagnostics-export.test.ts tests/projects-routes.test.ts tests/runs.test.ts tests/amr-acp-integration.test.ts tests/runtimes/env-and-detection.test.ts tests/runtimes/resolve-model.test.ts tests/runtimes/agent-args.test.ts --testNamePattern "grok-build inlines the prompt as -p and never falls back to stdin sentinels" +``` + +```bash +pnpm exec vitest run tests/integrations/vela.routes.test.ts tests/runs.test.ts --testNamePattern "uses the same Settings-configured AMR env for login and subsequent status reads|cancels a queued run immediately without waiting for child process shutdown" +``` + +```bash +pnpm exec vitest run tests/chat-run-artifact-quiet-period.test.ts tests/runtimes/env-and-detection.test.ts --testNamePattern "detectAgents prefers configured AMR live models over stale fallback defaults|returns succeeded when the watchdog-initiated quiet-period SIGTERM fires|returns failed when SIGTERM/SIGKILL arrive but no quiet-period shutdown was requested" +``` + +```bash +pnpm exec vitest run tests/chat-route.test.ts --testNamePattern "reuses an existing assistant message row instead of creating a duplicate when assistantMessageId is supplied" ``` ## 这批 daemon 补测当前没有覆盖的点 @@ -409,13 +497,13 @@ pnpm exec vitest run tests/chat-route.test.ts tests/diagnostics-export.test.ts t 下面这些仍然值得继续补，但这轮没有为了追求数量硬塞进去： 1. AMR / agent 运行结束态收敛 - - 例如“工作完成但没有 terminal event，最后被 watchdog 打成 failed” - - 例如“有有效产物但收尾阶段卡住”的 terminal-state 修正 + - quiet-period / watchdog 的 close-status 分流已经有专用测试文件锁住 + - 更深一层仍可继续补“真实 HTTP live-artifact create 后收尾成功”的整链回归 2. AMR auth / model discovery 的更完整契约 - - 例如 auth probe 与真实 launch path / env 必须同源 - - 例如 live models 成功时不能回退到假默认模型 + - 已补 login/status 同源配置，以及 `/api/agents` 的 live catalog 不回退假默认 + - 更深一层仍可继续补 live model discovery 与运行时 launch path 的完全同源 -3. queued / retry 的持久化语义 - - 前端行为已覆盖 - - daemon 侧仍可继续锁住 message 关联和队列启动顺序 +3. queued / retry 的更深层持久化语义 + - 前端行为已覆盖，daemon 侧已补 queued run 取消终态 + - 仍可继续锁住 retry 的 message 关联和队列启动顺序