@cursor/july 0.1.19 → 0.1.21

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (217) hide show
  1. package/AGENTS.md +10 -0
  2. package/dist/bin/agent-serve.js +0 -0
  3. package/dist/channels/github/instrument.d.ts +20 -0
  4. package/dist/channels/github/instrument.d.ts.map +1 -0
  5. package/dist/channels/slack/api.d.ts +2 -0
  6. package/dist/channels/slack/api.d.ts.map +1 -1
  7. package/dist/channels/slack/api.js +3 -0
  8. package/dist/channels/slack/bot-mentions.d.ts +106 -0
  9. package/dist/channels/slack/bot-mentions.d.ts.map +1 -0
  10. package/dist/channels/slack/bot-mentions.js +243 -0
  11. package/dist/channels/slack/dispatch.d.ts +6 -0
  12. package/dist/channels/slack/dispatch.d.ts.map +1 -1
  13. package/dist/channels/slack/dispatch.js +24 -3
  14. package/dist/channels/slack/inbound.d.ts +10 -1
  15. package/dist/channels/slack/inbound.d.ts.map +1 -1
  16. package/dist/channels/slack/inbound.js +12 -4
  17. package/dist/channels/slack/index.d.ts +1 -0
  18. package/dist/channels/slack/index.d.ts.map +1 -1
  19. package/dist/channels/slack/index.js +1 -0
  20. package/dist/channels/slack/slack-channel.d.ts.map +1 -1
  21. package/dist/channels/slack/slack-channel.js +61 -22
  22. package/dist/channels/slack/types.d.ts +58 -0
  23. package/dist/channels/slack/types.d.ts.map +1 -1
  24. package/dist/docs/404.html +1 -1
  25. package/dist/docs/ab.html +2 -2
  26. package/dist/docs/assets/{app.CrsWMchO.js → app.jDxLzWv4.js} +1 -1
  27. package/dist/docs/assets/chunks/@localSearchIndexroot.DoJHJjqF.js +1 -0
  28. package/dist/docs/assets/chunks/{VPLocalSearchBox.D1JqzSh8.js → VPLocalSearchBox.Y6bDR1-a.js} +1 -1
  29. package/dist/docs/assets/chunks/{theme.DaBvZYwl.js → theme.CLazCWlJ.js} +2 -2
  30. package/dist/docs/building-with-agents.html +2 -2
  31. package/dist/docs/concepts.html +2 -2
  32. package/dist/docs/deployment.html +2 -2
  33. package/dist/docs/evals.html +2 -2
  34. package/dist/docs/example-agents/approval-buddy.html +2 -2
  35. package/dist/docs/example-agents/benny.html +2 -2
  36. package/dist/docs/example-agents/bugbot.html +2 -2
  37. package/dist/docs/example-agents/codebase-wiki.html +2 -2
  38. package/dist/docs/example-agents/codeowners-review.html +2 -2
  39. package/dist/docs/example-agents/concierge.html +2 -2
  40. package/dist/docs/example-agents/fsd.html +2 -2
  41. package/dist/docs/example-agents/index.html +2 -2
  42. package/dist/docs/example-agents/knowledge-base.html +2 -2
  43. package/dist/docs/example-agents/oncall.html +2 -2
  44. package/dist/docs/example-agents/security-reviewer.html +2 -2
  45. package/dist/docs/example-agents/slack-agent.html +2 -2
  46. package/dist/docs/example-agents/weather-agent.html +2 -2
  47. package/dist/docs/guides/agent-to-agent.html +2 -2
  48. package/dist/docs/guides/cloud-runtime.html +2 -2
  49. package/dist/docs/guides/github.html +2 -2
  50. package/dist/docs/guides/human-in-the-loop.html +2 -2
  51. package/dist/docs/guides/mcp-oauth.html +2 -2
  52. package/dist/docs/guides/slack.html +2 -2
  53. package/dist/docs/guides/webhooks.html +2 -2
  54. package/dist/docs/hillclimbing.html +2 -2
  55. package/dist/docs/index.html +2 -2
  56. package/dist/docs/quickstart.html +2 -2
  57. package/dist/docs/reference/agent-config.html +2 -2
  58. package/dist/docs/reference/channels.html +2 -2
  59. package/dist/docs/reference/cli.html +2 -2
  60. package/dist/docs/reference/connections.html +2 -2
  61. package/dist/docs/reference/hooks.html +2 -2
  62. package/dist/docs/reference/http-api.html +2 -2
  63. package/dist/docs/reference/instructions.html +2 -2
  64. package/dist/docs/reference/playground.html +2 -2
  65. package/dist/docs/reference/project-layout.html +2 -2
  66. package/dist/docs/reference/prompt.html +2 -2
  67. package/dist/docs/reference/schedules.html +2 -2
  68. package/dist/docs/reference/sessions.html +2 -2
  69. package/dist/docs/reference/skills.html +2 -2
  70. package/dist/docs/reference/subagents.html +2 -2
  71. package/dist/docs/reference/tools.html +2 -2
  72. package/dist/docs/scaffolding-agents.html +2 -2
  73. package/dist/docs/storage.html +2 -2
  74. package/dist/docs/troubleshooting.html +2 -2
  75. package/dist/internal/json-dir-store.d.ts +32 -0
  76. package/dist/internal/json-dir-store.d.ts.map +1 -0
  77. package/dist/internal/json-dir-store.js +100 -0
  78. package/dist/internal/resolved-connections.d.ts +1 -1
  79. package/dist/internal/resolved-connections.d.ts.map +1 -1
  80. package/dist/internal/resolved-connections.js +10 -0
  81. package/dist/internal/session-engine.d.ts +4 -1
  82. package/dist/internal/session-engine.d.ts.map +1 -1
  83. package/dist/internal/session-engine.js +13 -2
  84. package/dist/playground/assets/index-CjOQ4hN9.css +1 -0
  85. package/dist/playground/assets/{index-C2SU2xV5.js → index-dshZQJCp.js} +46 -46
  86. package/dist/playground/index.html +2 -2
  87. package/dist/types.d.ts +7 -0
  88. package/dist/types.d.ts.map +1 -1
  89. package/dist/types.js +9 -0
  90. package/package.json +25 -25
  91. package/skills/create-agent/SKILL.md +36 -4
  92. package/skills/evals/SKILL.md +3 -0
  93. package/skills/framework-map/SKILL.md +14 -0
  94. package/skills/github/SKILL.md +6 -1
  95. package/skills/hillclimb/SKILL.md +28 -7
  96. package/src/bin/agent-serve.version.test.ts +62 -0
  97. package/src/channels/github/api.test.ts +64 -0
  98. package/src/channels/github/auth.test.ts +105 -0
  99. package/src/channels/github/cursor-account.test.ts +204 -0
  100. package/src/channels/github/forward.test.ts +457 -0
  101. package/src/channels/github/github.test.ts +937 -0
  102. package/src/channels/github/replay.test.ts +179 -0
  103. package/src/channels/slack/api.post-message.test.ts +148 -0
  104. package/src/channels/slack/api.ts +5 -0
  105. package/src/channels/slack/approvals.test.ts +328 -0
  106. package/src/channels/slack/block-actions.test.ts +452 -0
  107. package/src/channels/slack/bot-mentions.test.ts +217 -0
  108. package/src/channels/slack/bot-mentions.ts +315 -0
  109. package/src/channels/slack/channel-watch.test.ts +363 -0
  110. package/src/channels/slack/cursor-account.test.ts +253 -0
  111. package/src/channels/slack/defaults.final-post.test.ts +182 -0
  112. package/src/channels/slack/dispatch.test.ts +795 -0
  113. package/src/channels/slack/dispatch.ts +27 -1
  114. package/src/channels/slack/eval-directive.test.ts +273 -0
  115. package/src/channels/slack/inbound.ts +25 -4
  116. package/src/channels/slack/index.ts +1 -0
  117. package/src/channels/slack/message-body.test.ts +54 -0
  118. package/src/channels/slack/nudge-store.test.ts +143 -0
  119. package/src/channels/slack/slack-channel.ts +66 -8
  120. package/src/channels/slack/slack.test.ts +391 -0
  121. package/src/channels/slack/stop.test.ts +23 -0
  122. package/src/channels/slack/thread-context.test.ts +202 -0
  123. package/src/channels/slack/types.ts +60 -0
  124. package/src/evals/assertions.test.ts +580 -0
  125. package/src/evals/expect.test.ts +144 -0
  126. package/src/evals/judge.test.ts +181 -0
  127. package/src/evals/loaders.test.ts +132 -0
  128. package/src/evals/matchers.test.ts +95 -0
  129. package/src/evals/reporters.test.ts +303 -0
  130. package/src/evals/run-facts.test.ts +259 -0
  131. package/src/internal/ab-snapshot.test.ts +325 -0
  132. package/src/internal/approval-gate.test.ts +49 -0
  133. package/src/internal/approvals.integration.test.ts +383 -0
  134. package/src/internal/authored-loaders.test.ts +31 -0
  135. package/src/internal/builtin-tools/reminders.test.ts +201 -0
  136. package/src/internal/channel-route-schema.test.ts +294 -0
  137. package/src/internal/chat-attach.test.ts +262 -0
  138. package/src/internal/cli-deploy.test.ts +1991 -0
  139. package/src/internal/cli-mcp.test.ts +789 -0
  140. package/src/internal/cli-skills.test.ts +133 -0
  141. package/src/internal/cli-slack.test.ts +1647 -0
  142. package/src/internal/cloud-merge.test.ts +74 -0
  143. package/src/internal/cron.test.ts +22 -0
  144. package/src/internal/cursor/account-mcp.test.ts +807 -0
  145. package/src/internal/cursor/backend-client.test.ts +591 -0
  146. package/src/internal/cursor/credentials.test.ts +351 -0
  147. package/src/internal/cursor/github-credentials.test.ts +136 -0
  148. package/src/internal/cursor-account-mcp-auth.test.ts +310 -0
  149. package/src/internal/cursor-account.integration.test.ts +441 -0
  150. package/src/internal/cursor-event-relay.test.ts +746 -0
  151. package/src/internal/cursor-github-credentials.integration.test.ts +271 -0
  152. package/src/internal/cursor-slack-relay.test.ts +525 -0
  153. package/src/internal/deploy-source.test.ts +111 -0
  154. package/src/internal/discovery.builtin-tools.test.ts +94 -0
  155. package/src/internal/discovery.concurrency.test.ts +60 -0
  156. package/src/internal/discovery.cursor-account.test.ts +133 -0
  157. package/src/internal/discovery.cwd.test.ts +83 -0
  158. package/src/internal/discovery.hosting.test.ts +80 -0
  159. package/src/internal/discovery.identity.test.ts +44 -0
  160. package/src/internal/docs-site.test.ts +66 -0
  161. package/src/internal/duration.test.ts +29 -0
  162. package/src/internal/eval-judge-model.test.ts +187 -0
  163. package/src/internal/eval-run-store.cancel.test.ts +142 -0
  164. package/src/internal/eval-run-store.storage.test.ts +211 -0
  165. package/src/internal/eval-runner.http.test.ts +403 -0
  166. package/src/internal/eval-runner.run.test.ts +928 -0
  167. package/src/internal/evals-client.test.ts +307 -0
  168. package/src/internal/event-mapper.test.ts +243 -0
  169. package/src/internal/github-fanout.test.ts +213 -0
  170. package/src/internal/handleAgentServeTrigger.test.ts +179 -0
  171. package/src/internal/host-kv.test.ts +82 -0
  172. package/src/internal/host-platforms.test.ts +126 -0
  173. package/src/internal/http-channel.test.ts +402 -0
  174. package/src/internal/init-project.test.ts +269 -0
  175. package/src/internal/install-cursor-skills.test.ts +262 -0
  176. package/src/internal/local-env.test.ts +100 -0
  177. package/src/internal/log-ring.test.ts +31 -0
  178. package/src/internal/logs-client.test.ts +350 -0
  179. package/src/internal/mcp-endpoint.test.ts +436 -0
  180. package/src/internal/mcp-host.test.ts +298 -0
  181. package/src/internal/mcp-oauth.test.ts +148 -0
  182. package/src/internal/net.test.ts +17 -0
  183. package/src/internal/peer-connections.test.ts +128 -0
  184. package/src/internal/peer-mcp.integration.test.ts +289 -0
  185. package/src/internal/playground/toolchain.test.ts +53 -0
  186. package/src/internal/playground-cli.test.ts +187 -0
  187. package/src/internal/playground-proxy.test.ts +376 -0
  188. package/src/internal/prompt-context.integration.test.ts +232 -0
  189. package/src/internal/prompt-context.test.ts +127 -0
  190. package/src/internal/reminder-runner.test.ts +390 -0
  191. package/src/internal/reminder-store.test.ts +53 -0
  192. package/src/internal/request-headers.test.ts +27 -0
  193. package/src/internal/resolve-prod-target.test.ts +787 -0
  194. package/src/internal/resolved-connections.test.ts +295 -0
  195. package/src/internal/resolved-connections.ts +18 -1
  196. package/src/internal/router.test.ts +57 -0
  197. package/src/internal/sdk-runner.test.ts +290 -0
  198. package/src/internal/session-engine.coalesce.test.ts +169 -0
  199. package/src/internal/session-engine.concurrency.test.ts +250 -0
  200. package/src/internal/session-engine.host-oauth-mcp.test.ts +110 -0
  201. package/src/internal/session-engine.interrupt.test.ts +577 -0
  202. package/src/internal/session-engine.storage.test.ts +547 -0
  203. package/src/internal/session-engine.ts +15 -1
  204. package/src/internal/session-urls.test.ts +28 -0
  205. package/src/internal/sessions-client.test.ts +518 -0
  206. package/src/internal/storage-coordinator.test.ts +517 -0
  207. package/src/internal/tool-call.test.ts +458 -0
  208. package/src/internal/tool-result.test.ts +52 -0
  209. package/src/internal/trajectory.approvals.test.ts +83 -0
  210. package/src/internal/trajectory.subagents.test.ts +198 -0
  211. package/src/internal/turn-governor.test.ts +137 -0
  212. package/src/internal/update-check.test.ts +485 -0
  213. package/src/internal/workspace.test.ts +81 -0
  214. package/src/storage-backends/cursor-hosted.test.ts +121 -0
  215. package/src/types.ts +12 -0
  216. package/dist/docs/assets/chunks/@localSearchIndexroot.BnHRjfoe.js +0 -1
  217. package/dist/playground/assets/index-CidizGZv.css +0 -1
@@ -0,0 +1,458 @@
1
+ import { existsSync } from "node:fs";
2
+ import { mkdtemp, rm } from "node:fs/promises";
3
+ import { tmpdir } from "node:os";
4
+ import { join } from "node:path";
5
+ import { afterEach, describe, expect, it } from "vitest";
6
+ import { z } from "zod";
7
+ import { httpChannel } from "../channels.js";
8
+ import { defineTool } from "../tools.js";
9
+ import {
10
+ type AgentProject,
11
+ DIRECT_TOOL_CALL_CHANNEL_ID,
12
+ type DiscoveredTool,
13
+ type ToolDefinition,
14
+ } from "../types.js";
15
+ import { toolInputJsonSchema } from "./json-schema.js";
16
+ import type { AgentRunner } from "./sdk-runner.js";
17
+ import { type AgentMount, startServer } from "./server.js";
18
+ import {
19
+ SessionBusyError,
20
+ SessionEngine,
21
+ ToolInputError,
22
+ ToolNotDirectlyCallableError,
23
+ UnknownSessionError,
24
+ UnknownToolError,
25
+ } from "./session-engine.js";
26
+ import { buildTrajectory } from "./trajectory.js";
27
+
28
+ const echoTool = defineTool({
29
+ description: "Echo a message back.",
30
+ inputSchema: z.object({ message: z.string() }),
31
+ execute: ({ message }, ctx) => ({
32
+ echoed: message,
33
+ channelId: ctx.session.channelId,
34
+ sessionId: ctx.session.id,
35
+ workspaceDir: ctx.workspaceDir,
36
+ toolCallId: ctx.toolCallId,
37
+ hasAgentsMd: existsSync(join(ctx.workspaceDir, "AGENTS.md")),
38
+ }),
39
+ });
40
+
41
+ const throwingTool = defineTool({
42
+ description: "Always throws.",
43
+ execute: () => {
44
+ throw new Error("boom");
45
+ },
46
+ });
47
+
48
+ const scriptTool = defineTool({
49
+ description: "Agent-side script tool.",
50
+ execution: "agent",
51
+ script: "#!/usr/bin/env bash\ncat\n",
52
+ });
53
+
54
+ function discovered(name: string, definition: ToolDefinition): DiscoveredTool {
55
+ return {
56
+ name,
57
+ description: definition.description,
58
+ inputSchema: toolInputJsonSchema(definition.inputSchema),
59
+ execution: definition.execution ?? "server",
60
+ needsApproval: definition.needsApproval !== undefined,
61
+ ...(definition.script === undefined ? {} : { script: definition.script }),
62
+ definition,
63
+ };
64
+ }
65
+
66
+ function testProject(rootDir: string): AgentProject {
67
+ return {
68
+ rootDir,
69
+ agentDir: join(rootDir, "agent"),
70
+ name: "tool-call-test",
71
+ agent: {
72
+ name: "tool-call-test",
73
+ runtime: "local",
74
+ instructions: "Test agent instructions.",
75
+ tools: [
76
+ discovered("echo", echoTool),
77
+ discovered("throws", throwingTool),
78
+ discovered("script_tool", scriptTool),
79
+ ],
80
+ skills: [],
81
+ connections: [],
82
+ subagents: [],
83
+ seedFiles: [],
84
+ },
85
+ httpChannel: httpChannel(),
86
+ channels: [],
87
+ schedules: [],
88
+ hooks: [],
89
+ abs: [],
90
+ diagnostics: [],
91
+ };
92
+ }
93
+
94
+ /** Runner that never runs a model turn — direct calls must not need one. */
95
+ const noTurnRunner: AgentRunner = {
96
+ runTurn: () => {
97
+ throw new Error("runTurn must not be called for direct tool calls");
98
+ },
99
+ };
100
+
101
+ const cleanups: Array<() => Promise<void>> = [];
102
+
103
+ afterEach(async () => {
104
+ while (cleanups.length > 0) {
105
+ await cleanups.pop()?.();
106
+ }
107
+ });
108
+
109
+ async function makeEngine(): Promise<SessionEngine> {
110
+ const stateRoot = await mkdtemp(join(tmpdir(), "agent-serve-toolcall-"));
111
+ cleanups.push(() => rm(stateRoot, { recursive: true, force: true }));
112
+ return new SessionEngine({
113
+ project: testProject(stateRoot),
114
+ stateRoot,
115
+ runner: noTurnRunner,
116
+ });
117
+ }
118
+
119
+ describe("SessionEngine.callTool", () => {
120
+ it("runs a server tool deterministically with a scratch context", async () => {
121
+ const engine = await makeEngine();
122
+ const outcome = await engine.callTool("echo", { message: "hi" });
123
+ expect(outcome.isError).toBe(false);
124
+ expect(outcome.toolName).toBe("echo");
125
+ expect(outcome.callId).toMatch(/^tool_echo_[0-9a-f]{16}$/);
126
+ const result = outcome.result as Record<string, string | boolean>;
127
+ expect(result.echoed).toBe("hi");
128
+ expect(result.channelId).toBe(DIRECT_TOOL_CALL_CHANNEL_ID);
129
+ expect(result.workspaceDir).toContain(join("tool-calls", outcome.callId));
130
+ expect(result.toolCallId).toBe(outcome.callId);
131
+ // Scratch workspaces are materialized like a session workspace…
132
+ expect(result.hasAgentsMd).toBe(true);
133
+ // …and removed once the call returns.
134
+ expect(existsSync(result.workspaceDir as string)).toBe(false);
135
+ expect(outcome.sessionId).toBeUndefined();
136
+ });
137
+
138
+ it("validates input against the zod schema before executing", async () => {
139
+ const engine = await makeEngine();
140
+ await expect(engine.callTool("echo", { message: 42 })).rejects.toThrow(
141
+ ToolInputError
142
+ );
143
+ });
144
+
145
+ it("rejects unknown tools with the available server tools", async () => {
146
+ const engine = await makeEngine();
147
+ const error = await engine.callTool("nope").catch((e: unknown) => e);
148
+ expect(error).toBeInstanceOf(UnknownToolError);
149
+ expect((error as UnknownToolError).availableToolNames).toEqual([
150
+ "echo",
151
+ "throws",
152
+ ]);
153
+ });
154
+
155
+ it("rejects agent-execution tools", async () => {
156
+ const engine = await makeEngine();
157
+ await expect(engine.callTool("script_tool", {})).rejects.toThrow(
158
+ ToolNotDirectlyCallableError
159
+ );
160
+ });
161
+
162
+ it("reports a throwing tool body with the model-path error envelope", async () => {
163
+ const engine = await makeEngine();
164
+ const outcome = await engine.callTool("throws");
165
+ expect(outcome.isError).toBe(true);
166
+ expect(outcome.result).toEqual({
167
+ content: [{ type: "text", text: 'Tool "throws" failed: boom' }],
168
+ isError: true,
169
+ });
170
+ expect(outcome.errorMessage).toBe('Tool "throws" failed: boom');
171
+ });
172
+
173
+ it("rejects unknown sessions", async () => {
174
+ const engine = await makeEngine();
175
+ await expect(
176
+ engine.callTool("echo", { message: "hi" }, { sessionId: "ses_missing" })
177
+ ).rejects.toThrow(UnknownSessionError);
178
+ });
179
+ });
180
+
181
+ describe("SessionEngine.callTool bound to a session", () => {
182
+ it("uses the session context and records the call on its event stream", async () => {
183
+ const engine = await makeEngine();
184
+ // Create a session record directly (no turn) so the runner stays unused.
185
+ const record = {
186
+ sessionId: "ses_toolcall_test",
187
+ channelId: "http",
188
+ mode: "chat" as const,
189
+ continuationKey: null,
190
+ auth: null,
191
+ turnCount: 0,
192
+ createdAt: new Date().toISOString(),
193
+ updatedAt: new Date().toISOString(),
194
+ };
195
+ await engine.sessions.save(record);
196
+
197
+ const outcome = await engine.callTool(
198
+ "echo",
199
+ { message: "in-session" },
200
+ { sessionId: record.sessionId }
201
+ );
202
+ expect(outcome.isError).toBe(false);
203
+ expect(outcome.sessionId).toBe(record.sessionId);
204
+ const result = outcome.result as Record<string, string>;
205
+ expect(result.sessionId).toBe(record.sessionId);
206
+ expect(result.channelId).toBe("http");
207
+ expect(result.workspaceDir).toBe(
208
+ engine.sessions.workspaceDir(record.sessionId)
209
+ );
210
+ expect(result.hasAgentsMd).toBe(true);
211
+
212
+ const events = await engine.listEvents(record.sessionId);
213
+ const requested = events.find((e) => e.type === "actions.requested");
214
+ const resulted = events.find((e) => e.type === "action.result");
215
+ expect(requested?.turnId).toBe(outcome.callId);
216
+ expect(requested?.data).toEqual({
217
+ calls: [
218
+ {
219
+ callId: outcome.callId,
220
+ toolName: "echo",
221
+ args: { message: "in-session" },
222
+ },
223
+ ],
224
+ });
225
+ expect(resulted?.turnId).toBe(outcome.callId);
226
+ expect(resulted?.data).toMatchObject({
227
+ callId: outcome.callId,
228
+ toolName: "echo",
229
+ isError: false,
230
+ });
231
+
232
+ // The per-call turnId keeps direct calls visible in trajectories.
233
+ const trajectory = buildTrajectory(events);
234
+ const trajectoryCall = trajectory.turns
235
+ .flatMap((turn) => turn.toolCalls)
236
+ .find((call) => call.callId === outcome.callId);
237
+ expect(trajectoryCall).toMatchObject({ toolName: "echo", isError: false });
238
+ });
239
+
240
+ it("rejects direct calls while a model turn is running", async () => {
241
+ const stateRoot = await mkdtemp(join(tmpdir(), "agent-serve-toolcall-"));
242
+ cleanups.push(() => rm(stateRoot, { recursive: true, force: true }));
243
+ let releaseTurn: () => void = () => {};
244
+ const gate = new Promise<void>((resolve) => {
245
+ releaseTurn = resolve;
246
+ });
247
+ const blockingRunner: AgentRunner = {
248
+ runTurn: async () => {
249
+ await gate;
250
+ return { status: "finished" as const, result: "done" };
251
+ },
252
+ };
253
+ const engine = new SessionEngine({
254
+ project: testProject(stateRoot),
255
+ stateRoot,
256
+ runner: blockingRunner,
257
+ });
258
+ cleanups.push(() => engine.close());
259
+
260
+ const session = await engine.send("http", "hold the session", {
261
+ auth: null,
262
+ });
263
+ await expect(
264
+ engine.callTool("echo", { message: "hi" }, { sessionId: session.id })
265
+ ).rejects.toThrow(SessionBusyError);
266
+ releaseTurn();
267
+ await session.waitForCompletion();
268
+ });
269
+ });
270
+
271
+ describe("POST /v1/tools/:toolName", () => {
272
+ async function startTestServer(): Promise<{ url: string }> {
273
+ const stateRoot = await mkdtemp(join(tmpdir(), "agent-serve-toolhttp-"));
274
+ cleanups.push(() => rm(stateRoot, { recursive: true, force: true }));
275
+ const mounts: AgentMount[] = [
276
+ { slug: "", project: testProject(stateRoot) },
277
+ ];
278
+ const handle = await startServer(mounts, {
279
+ port: 0,
280
+ stateRoot,
281
+ playground: false,
282
+ schedules: false,
283
+ logger: () => {},
284
+ });
285
+ cleanups.push(() => handle.close());
286
+ return { url: handle.url };
287
+ }
288
+
289
+ it("calls a tool over HTTP and returns the normalized result", async () => {
290
+ const { url } = await startTestServer();
291
+ const response = await fetch(`${url}/v1/tools/echo`, {
292
+ method: "POST",
293
+ headers: { "content-type": "application/json" },
294
+ body: JSON.stringify({ input: { message: "over http" } }),
295
+ });
296
+ expect(response.status).toBe(200);
297
+ const body = (await response.json()) as {
298
+ ok: boolean;
299
+ toolName: string;
300
+ isError: boolean;
301
+ result: { echoed: string };
302
+ };
303
+ expect(body.ok).toBe(true);
304
+ expect(body.toolName).toBe("echo");
305
+ expect(body.isError).toBe(false);
306
+ expect(body.result.echoed).toBe("over http");
307
+ });
308
+
309
+ it("returns 404 with available tools for unknown tools", async () => {
310
+ const { url } = await startTestServer();
311
+ const response = await fetch(`${url}/v1/tools/nope`, {
312
+ method: "POST",
313
+ headers: { "content-type": "application/json" },
314
+ body: JSON.stringify({}),
315
+ });
316
+ expect(response.status).toBe(404);
317
+ const body = (await response.json()) as {
318
+ error: string;
319
+ availableTools: string[];
320
+ };
321
+ expect(body.error).toBe("unknown_tool");
322
+ expect(body.availableTools).toEqual(["echo", "throws"]);
323
+ });
324
+
325
+ it("returns 400 on schema-invalid input", async () => {
326
+ const { url } = await startTestServer();
327
+ const response = await fetch(`${url}/v1/tools/echo`, {
328
+ method: "POST",
329
+ headers: { "content-type": "application/json" },
330
+ body: JSON.stringify({ input: { message: 42 } }),
331
+ });
332
+ expect(response.status).toBe(400);
333
+ const body = (await response.json()) as { error: string };
334
+ expect(body.error).toBe("invalid_tool_input");
335
+ });
336
+
337
+ it("returns 400 for agent-execution tools", async () => {
338
+ const { url } = await startTestServer();
339
+ const response = await fetch(`${url}/v1/tools/script_tool`, {
340
+ method: "POST",
341
+ headers: { "content-type": "application/json" },
342
+ body: JSON.stringify({}),
343
+ });
344
+ expect(response.status).toBe(400);
345
+ const body = (await response.json()) as { error: string };
346
+ expect(body.error).toBe("tool_not_callable");
347
+ });
348
+
349
+ it("returns ok: false with the error envelope when the tool body throws", async () => {
350
+ const { url } = await startTestServer();
351
+ const response = await fetch(`${url}/v1/tools/throws`, {
352
+ method: "POST",
353
+ headers: { "content-type": "application/json" },
354
+ body: JSON.stringify({}),
355
+ });
356
+ expect(response.status).toBe(200);
357
+ const body = (await response.json()) as {
358
+ ok: boolean;
359
+ isError: boolean;
360
+ result: unknown;
361
+ };
362
+ expect(body.ok).toBe(false);
363
+ expect(body.isError).toBe(true);
364
+ expect(body.result).toEqual({
365
+ content: [{ type: "text", text: 'Tool "throws" failed: boom' }],
366
+ isError: true,
367
+ });
368
+ });
369
+
370
+ it("returns 400 for a malformed or missing JSON body", async () => {
371
+ const { url } = await startTestServer();
372
+ const malformed = await fetch(`${url}/v1/tools/echo`, {
373
+ method: "POST",
374
+ headers: { "content-type": "application/json" },
375
+ body: "{not json",
376
+ });
377
+ expect(malformed.status).toBe(400);
378
+ expect(((await malformed.json()) as { error: string }).error).toBe(
379
+ "invalid_request"
380
+ );
381
+
382
+ const nonObject = await fetch(`${url}/v1/tools/echo`, {
383
+ method: "POST",
384
+ headers: { "content-type": "application/json" },
385
+ body: JSON.stringify([1, 2]),
386
+ });
387
+ expect(nonObject.status).toBe(400);
388
+ });
389
+
390
+ it("returns 400 for a non-string sessionId", async () => {
391
+ const { url } = await startTestServer();
392
+ const response = await fetch(`${url}/v1/tools/echo`, {
393
+ method: "POST",
394
+ headers: { "content-type": "application/json" },
395
+ body: JSON.stringify({ input: { message: "hi" }, sessionId: 42 }),
396
+ });
397
+ expect(response.status).toBe(400);
398
+ expect(((await response.json()) as { error: string }).error).toBe(
399
+ "invalid_request"
400
+ );
401
+ });
402
+
403
+ it("returns 404 for an unknown sessionId", async () => {
404
+ const { url } = await startTestServer();
405
+ const response = await fetch(`${url}/v1/tools/echo`, {
406
+ method: "POST",
407
+ headers: { "content-type": "application/json" },
408
+ body: JSON.stringify({
409
+ input: { message: "hi" },
410
+ sessionId: "ses_missing",
411
+ }),
412
+ });
413
+ expect(response.status).toBe(404);
414
+ const body = (await response.json()) as { error: string };
415
+ expect(body.error).toBe("session_not_found");
416
+ });
417
+ });
418
+
419
+ describe("multi-agent host /v1/health", () => {
420
+ it("exposes host liveness at /v1/health alongside per-slug health", async () => {
421
+ const stateRoot = await mkdtemp(join(tmpdir(), "agent-serve-multihealth-"));
422
+ cleanups.push(() => rm(stateRoot, { recursive: true, force: true }));
423
+ const mounts: AgentMount[] = [
424
+ { slug: "alpha", project: testProject(stateRoot) },
425
+ { slug: "beta", project: testProject(stateRoot) },
426
+ ];
427
+ const handle = await startServer(mounts, {
428
+ port: 0,
429
+ stateRoot,
430
+ playground: false,
431
+ schedules: false,
432
+ logger: () => {},
433
+ });
434
+ cleanups.push(() => handle.close());
435
+
436
+ const host = await fetch(`${handle.url}/v1/health`);
437
+ expect(host.status).toBe(200);
438
+ const hostBody = (await host.json()) as {
439
+ ok: boolean;
440
+ agents: Array<{ slug: string; name: string }>;
441
+ };
442
+ expect(hostBody.ok).toBe(true);
443
+ expect(hostBody.agents.map((a) => a.slug).sort()).toEqual([
444
+ "alpha",
445
+ "beta",
446
+ ]);
447
+
448
+ const alpha = await fetch(`${handle.url}/alpha/v1/health`);
449
+ expect(alpha.status).toBe(200);
450
+ expect(((await alpha.json()) as { ok: boolean }).ok).toBe(true);
451
+
452
+ // Corp ECS health uses wget --spider (HEAD). GET-only routes must still
453
+ // answer 200 with an empty body so the container check does not fail.
454
+ const head = await fetch(`${handle.url}/v1/health`, { method: "HEAD" });
455
+ expect(head.status).toBe(200);
456
+ expect(await head.text()).toBe("");
457
+ });
458
+ });
@@ -0,0 +1,52 @@
1
+ import { describe, expect, it } from "vitest";
2
+ import { normalizeToolResult, toolCallErrorMessage } from "./tool-result.js";
3
+
4
+ describe("normalizeToolResult", () => {
5
+ it("passes strings through", () => {
6
+ expect(normalizeToolResult("ok")).toBe("ok");
7
+ });
8
+
9
+ it("maps undefined to the empty result", () => {
10
+ expect(normalizeToolResult(undefined)).toBe("");
11
+ });
12
+
13
+ it("drops undefined object fields so Struct.fromJson cannot throw", () => {
14
+ const result = normalizeToolResult({
15
+ kept: 1,
16
+ dropped: undefined,
17
+ nested: [{ action: undefined, status: "OPEN" }],
18
+ } as never);
19
+ expect(result).toEqual({ kept: 1, nested: [{ status: "OPEN" }] });
20
+ });
21
+
22
+ it("keeps content envelopes intact", () => {
23
+ const envelope = {
24
+ content: [{ type: "text" as const, text: "hi" }],
25
+ isError: true,
26
+ };
27
+ expect(normalizeToolResult(envelope)).toEqual(envelope);
28
+ });
29
+ });
30
+
31
+ describe("toolCallErrorMessage", () => {
32
+ it("returns string results", () => {
33
+ expect(toolCallErrorMessage("boom")).toBe("boom");
34
+ });
35
+
36
+ it("joins text parts from an error envelope", () => {
37
+ expect(
38
+ toolCallErrorMessage({
39
+ content: [
40
+ { type: "text", text: "first" },
41
+ { type: "text", text: " second " },
42
+ ],
43
+ isError: true,
44
+ })
45
+ ).toBe("first\nsecond");
46
+ });
47
+
48
+ it("returns undefined for non-text results", () => {
49
+ expect(toolCallErrorMessage({ ok: true })).toBeUndefined();
50
+ expect(toolCallErrorMessage("")).toBeUndefined();
51
+ });
52
+ });
@@ -0,0 +1,83 @@
1
+ import { describe, expect, it } from "vitest";
2
+ import type { SessionEvent } from "../types.js";
3
+ import { buildTrajectory } from "./trajectory.js";
4
+
5
+ function ev(
6
+ partial: Omit<SessionEvent, "index" | "sessionId" | "at"> & {
7
+ index?: number;
8
+ }
9
+ ): SessionEvent {
10
+ return {
11
+ index: partial.index ?? 0,
12
+ sessionId: "sess_1",
13
+ at: "2026-01-01T00:00:00.000Z",
14
+ ...partial,
15
+ } as SessionEvent;
16
+ }
17
+
18
+ describe("buildTrajectory approvals", () => {
19
+ it("records approvalDecision on the tool call", () => {
20
+ const traj = buildTrajectory([
21
+ ev({ type: "session.started", data: { channelId: "http" } }),
22
+ ev({
23
+ type: "message.received",
24
+ turnId: "turn_1",
25
+ data: { text: "alert ops about NYC" },
26
+ }),
27
+ ev({
28
+ type: "actions.requested",
29
+ turnId: "turn_1",
30
+ data: {
31
+ calls: [
32
+ {
33
+ callId: "call_1",
34
+ toolName: "post_weather_alert",
35
+ args: { city: "NYC" },
36
+ },
37
+ ],
38
+ },
39
+ }),
40
+ ev({
41
+ type: "action.approval_requested",
42
+ turnId: "turn_1",
43
+ data: {
44
+ callId: "call_1",
45
+ toolName: "post_weather_alert",
46
+ args: { city: "NYC" },
47
+ },
48
+ }),
49
+ ev({
50
+ type: "action.approval_resolved",
51
+ turnId: "turn_1",
52
+ data: {
53
+ callId: "call_1",
54
+ toolName: "post_weather_alert",
55
+ decision: "approve",
56
+ },
57
+ }),
58
+ ev({
59
+ type: "action.result",
60
+ turnId: "turn_1",
61
+ data: {
62
+ callId: "call_1",
63
+ toolName: "post_weather_alert",
64
+ output: { posted: true },
65
+ isError: false,
66
+ },
67
+ }),
68
+ ev({
69
+ type: "turn.completed",
70
+ turnId: "turn_1",
71
+ data: {},
72
+ }),
73
+ ]);
74
+
75
+ expect(traj.turns[0]?.toolCalls[0]).toMatchObject({
76
+ callId: "call_1",
77
+ toolName: "post_weather_alert",
78
+ approvalDecision: "approve",
79
+ isError: false,
80
+ output: { posted: true },
81
+ });
82
+ });
83
+ });