@vellumai/assistant 0.8.9-staging.1 → 0.8.9-staging.2

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 (62) hide show
  1. package/ARCHITECTURE.md +6 -6
  2. package/openapi.yaml +6 -7
  3. package/package.json +1 -1
  4. package/src/__tests__/agent-loop-exit-reason.test.ts +33 -18
  5. package/src/__tests__/app-builder-tool-scripts.test.ts +21 -0
  6. package/src/__tests__/app-executors.test.ts +132 -0
  7. package/src/__tests__/compactor-web-search-strip.test.ts +213 -0
  8. package/src/__tests__/context-overflow-reducer.test.ts +1 -1
  9. package/src/__tests__/conversation-agent-loop-inference-profile.test.ts +19 -16
  10. package/src/__tests__/conversation-agent-loop-overflow.test.ts +48 -29
  11. package/src/__tests__/conversation-agent-loop.test.ts +48 -96
  12. package/src/__tests__/conversation-history-web-search.test.ts +1 -1
  13. package/src/__tests__/conversation-provider-retry-repair.test.ts +52 -49
  14. package/src/__tests__/memory-retrieval-hook.test.ts +68 -30
  15. package/src/__tests__/pre-model-call-sanitize.test.ts +109 -0
  16. package/src/acp/__tests__/helpers/exec-file-stub.ts +12 -7
  17. package/src/acp/__tests__/prepare-agent-env.test.ts +0 -40
  18. package/src/acp/__tests__/session-manager-resume.test.ts +179 -27
  19. package/src/acp/auto-install.test.ts +130 -46
  20. package/src/acp/auto-install.ts +86 -31
  21. package/src/acp/prepare-agent-env.ts +8 -9
  22. package/src/acp/resolve-agent.test.ts +26 -123
  23. package/src/acp/resolve-agent.ts +26 -85
  24. package/src/acp/resume-hint.ts +1 -3
  25. package/src/acp/session-manager.ts +32 -20
  26. package/src/acp/types.ts +0 -8
  27. package/src/agent/loop.ts +42 -26
  28. package/src/config/bundled-skills/acp/SKILL.md +14 -14
  29. package/src/config/bundled-skills/acp/TOOLS.json +2 -2
  30. package/src/config/bundled-skills/app-builder/SKILL.md +3 -3
  31. package/src/config/bundled-skills/app-builder/TOOLS.json +43 -2
  32. package/src/config/bundled-skills/app-builder/tools/app-update.ts +18 -0
  33. package/src/config/bundled-tool-registry.ts +2 -0
  34. package/src/config/call-site-defaults.ts +0 -2
  35. package/src/config/feature-flag-registry.json +1 -1
  36. package/src/config/schemas/call-site-catalog.ts +0 -14
  37. package/src/config/schemas/llm.ts +0 -2
  38. package/src/context/compactor.ts +22 -4
  39. package/src/daemon/conversation-agent-loop.ts +15 -57
  40. package/src/daemon/conversation.ts +23 -3
  41. package/src/daemon/disk-pressure-policy.ts +0 -1
  42. package/src/daemon/lifecycle.ts +0 -7
  43. package/src/{daemon → plugins/defaults/compaction}/context-overflow-reducer.ts +7 -7
  44. package/src/plugins/defaults/compaction/manager-store.ts +13 -0
  45. package/src/plugins/defaults/memory-retrieval/hooks/post-compact.ts +6 -23
  46. package/src/plugins/defaults/memory-retrieval/hooks/user-prompt-submit-temp.ts +24 -25
  47. package/src/runtime/routes/__tests__/acp-routes.test.ts +71 -39
  48. package/src/runtime/routes/__tests__/plugins-routes.test.ts +29 -4
  49. package/src/runtime/routes/plugins-routes.ts +10 -9
  50. package/src/tools/acp/list-agents.test.ts +2 -2
  51. package/src/tools/acp/spawn.test.ts +103 -225
  52. package/src/tools/acp/spawn.ts +10 -120
  53. package/src/tools/apps/executors.ts +153 -42
  54. package/src/proactive-artifact/aux-message-injector.ts +0 -97
  55. package/src/proactive-artifact/decision.test.ts +0 -226
  56. package/src/proactive-artifact/decision.ts +0 -165
  57. package/src/proactive-artifact/index.ts +0 -7
  58. package/src/proactive-artifact/job.test.ts +0 -962
  59. package/src/proactive-artifact/job.ts +0 -372
  60. package/src/proactive-artifact/message-copy.ts +0 -58
  61. package/src/proactive-artifact/trigger-state.test.ts +0 -286
  62. package/src/proactive-artifact/trigger-state.ts +0 -123
@@ -68,6 +68,73 @@ export type ProxyResolver = (
68
68
  input: Record<string, unknown>,
69
69
  ) => Promise<ExecutorResult>;
70
70
 
71
+ /**
72
+ * Validate a `source_files` map (LLM input is not type-checked at runtime).
73
+ * Returns an error ExecutorResult when invalid, or null when absent/valid.
74
+ */
75
+ function validateSourceFiles(sourceFiles: unknown): ExecutorResult | null {
76
+ if (sourceFiles == null) {
77
+ return null;
78
+ }
79
+ if (typeof sourceFiles !== "object" || Array.isArray(sourceFiles)) {
80
+ return {
81
+ content: JSON.stringify({
82
+ error:
83
+ "source_files must be an object mapping relative file paths to string contents",
84
+ }),
85
+ isError: true,
86
+ };
87
+ }
88
+ for (const [key, val] of Object.entries(sourceFiles)) {
89
+ if (typeof val !== "string") {
90
+ return {
91
+ content: JSON.stringify({
92
+ error: `source_files["${key}"] must be a string, got ${typeof val}`,
93
+ }),
94
+ isError: true,
95
+ };
96
+ }
97
+ }
98
+ return null;
99
+ }
100
+
101
+ /**
102
+ * Resolve an app name when the model omits it. The preview card's title is
103
+ * "always include" guidance and almost always present, so it's the best
104
+ * fallback — a real, meaningful name — before a generic placeholder.
105
+ */
106
+ function resolveAppName(input: AppCreateInput): string {
107
+ const previewTitle =
108
+ input.preview && typeof input.preview.title === "string"
109
+ ? input.preview.title
110
+ : undefined;
111
+ for (const candidate of [input.name, previewTitle]) {
112
+ if (typeof candidate === "string" && candidate.trim() !== "") {
113
+ return candidate.trim();
114
+ }
115
+ }
116
+ return "New App";
117
+ }
118
+
119
+ /**
120
+ * Compile-result fields shared by the app_refresh and app_update responses.
121
+ * On failure the errors/warnings are surfaced so the agent can fix them.
122
+ */
123
+ function compileResultPayload(
124
+ compileResult: Awaited<ReturnType<typeof compileApp>>,
125
+ ): Record<string, unknown> {
126
+ return {
127
+ compiled: compileResult.ok,
128
+ ...(compileResult.ok
129
+ ? { compile_duration_ms: compileResult.durationMs }
130
+ : {
131
+ compile_errors: compileResult.errors,
132
+ compile_warnings: compileResult.warnings,
133
+ compile_duration_ms: compileResult.durationMs,
134
+ }),
135
+ };
136
+ }
137
+
71
138
  // ---------------------------------------------------------------------------
72
139
  // app_create
73
140
  // ---------------------------------------------------------------------------
@@ -92,7 +159,9 @@ export async function executeAppCreate(
92
159
  store: AppStore,
93
160
  proxyToolResolver?: ProxyResolver,
94
161
  ): Promise<ExecutorResult> {
95
- const name = input.name;
162
+ // The model sometimes omits a name; resolve a sensible one rather than
163
+ // erroring out so the build still succeeds. Users can rename via app_update.
164
+ const name = resolveAppName(input);
96
165
  const description = input.description;
97
166
  const schemaJson = input.schema_json ?? "{}";
98
167
 
@@ -121,39 +190,9 @@ export async function executeAppCreate(
121
190
  const autoOpen = input.auto_open !== false; // default true
122
191
  const preview = input.preview;
123
192
 
124
- // Validate required fields - LLM input is not type-checked at runtime
125
- if (typeof name !== "string" || name.trim() === "") {
126
- return {
127
- content: JSON.stringify({
128
- error: "name is required and must be a non-empty string",
129
- }),
130
- isError: true,
131
- };
132
- }
133
-
134
- if (input.source_files != null) {
135
- if (
136
- typeof input.source_files !== "object" ||
137
- Array.isArray(input.source_files)
138
- ) {
139
- return {
140
- content: JSON.stringify({
141
- error:
142
- "source_files must be an object mapping relative file paths to string contents",
143
- }),
144
- isError: true,
145
- };
146
- }
147
- for (const [key, val] of Object.entries(input.source_files)) {
148
- if (typeof val !== "string") {
149
- return {
150
- content: JSON.stringify({
151
- error: `source_files["${key}"] must be a string, got ${typeof val}`,
152
- }),
153
- isError: true,
154
- };
155
- }
156
- }
193
+ const sourceFilesError = validateSourceFiles(input.source_files);
194
+ if (sourceFilesError) {
195
+ return sourceFilesError;
157
196
  }
158
197
 
159
198
  // Extract icon from preview if provided - only persist emoji-like values,
@@ -356,14 +395,7 @@ export async function executeAppRefresh(
356
395
  refreshed: true,
357
396
  appId: updated.id,
358
397
  name: updated.name,
359
- compiled: compileResult.ok,
360
- ...(compileResult.ok
361
- ? { compile_duration_ms: compileResult.durationMs }
362
- : {
363
- compile_errors: compileResult.errors,
364
- compile_warnings: compileResult.warnings,
365
- compile_duration_ms: compileResult.durationMs,
366
- }),
398
+ ...compileResultPayload(compileResult),
367
399
  }),
368
400
  isError: false,
369
401
  };
@@ -379,6 +411,85 @@ export async function executeAppRefresh(
379
411
  };
380
412
  }
381
413
 
414
+ // ---------------------------------------------------------------------------
415
+ // app_update
416
+ // ---------------------------------------------------------------------------
417
+
418
+ export interface AppUpdateInput {
419
+ app_id: string;
420
+ name?: string;
421
+ description?: string;
422
+ schema_json?: string;
423
+ source_files?: Record<string, string>;
424
+ }
425
+
426
+ export async function executeAppUpdate(
427
+ input: AppUpdateInput,
428
+ store: AppStore,
429
+ ): Promise<ExecutorResult> {
430
+ const app = store.getApp(input.app_id);
431
+ if (!app) {
432
+ return {
433
+ content: JSON.stringify({ error: `App '${input.app_id}' not found` }),
434
+ isError: true,
435
+ };
436
+ }
437
+
438
+ const sourceFilesError = validateSourceFiles(input.source_files);
439
+ if (sourceFilesError) {
440
+ return sourceFilesError;
441
+ }
442
+
443
+ if (input.source_files) {
444
+ for (const [filePath, content] of Object.entries(input.source_files)) {
445
+ store.writeAppFile(input.app_id, filePath, content);
446
+ }
447
+ }
448
+
449
+ const updates: Partial<
450
+ Pick<AppDefinition, "name" | "description" | "schemaJson">
451
+ > = {};
452
+ if (typeof input.name === "string" && input.name.trim() !== "") {
453
+ updates.name = input.name.trim();
454
+ }
455
+ if (typeof input.description === "string") {
456
+ updates.description = input.description;
457
+ }
458
+ if (typeof input.schema_json === "string") {
459
+ updates.schemaJson = input.schema_json;
460
+ }
461
+
462
+ // An empty update still bumps updatedAt, triggering a client surface refresh.
463
+ const updated = store.updateApp(input.app_id, updates);
464
+
465
+ // Multifile apps recompile so the agent sees any errors from the edited
466
+ // source instead of serving a stale dist (mirrors app_refresh).
467
+ if (app.formatVersion === 2) {
468
+ const appDir = getAppDirPath(input.app_id);
469
+ const compileResult = await compileApp(appDir);
470
+ return {
471
+ content: JSON.stringify({
472
+ updated: true,
473
+ appId: updated.id,
474
+ name: updated.name,
475
+ description: updated.description,
476
+ ...compileResultPayload(compileResult),
477
+ }),
478
+ isError: false,
479
+ };
480
+ }
481
+
482
+ return {
483
+ content: JSON.stringify({
484
+ updated: true,
485
+ appId: updated.id,
486
+ name: updated.name,
487
+ description: updated.description,
488
+ }),
489
+ isError: false,
490
+ };
491
+ }
492
+
382
493
  // ---------------------------------------------------------------------------
383
494
  // app_generate_icon
384
495
  // ---------------------------------------------------------------------------
@@ -1,97 +0,0 @@
1
- /**
2
- * Auxiliary message injection for proactive artifacts.
3
- *
4
- * Injects an assistant message into a conversation without going through
5
- * the normal agent loop. Defers injection while the conversation is
6
- * actively processing to preserve chronological message ordering.
7
- */
8
-
9
- import { createAssistantMessage } from "../agent/message-types.js";
10
- import { findConversation } from "../daemon/conversation-registry.js";
11
- import {
12
- conversationMessagesSyncTag,
13
- SYNC_TAGS,
14
- } from "../daemon/message-types/sync.js";
15
- import { addMessage } from "../memory/conversation-crud.js";
16
- import type { BroadcastFn } from "../notifications/adapters/macos.js";
17
- import { getLogger } from "../util/logger.js";
18
-
19
- const log = getLogger("aux-message-injector");
20
- const IDLE_POLL_MS = 200;
21
- const IDLE_TIMEOUT_MS = 60_000;
22
-
23
- async function waitForIdle(conversationId: string): Promise<boolean> {
24
- const start = Date.now();
25
- while (Date.now() - start < IDLE_TIMEOUT_MS) {
26
- const conv = findConversation(conversationId);
27
- if (!conv || !conv.isProcessing()) return true;
28
- await new Promise((resolve) => setTimeout(resolve, IDLE_POLL_MS));
29
- }
30
- return false;
31
- }
32
-
33
- export async function injectAuxAssistantMessage(params: {
34
- conversationId: string;
35
- text: string;
36
- broadcastMessage: BroadcastFn;
37
- }): Promise<void> {
38
- const conv = findConversation(params.conversationId);
39
- if (conv?.isProcessing()) {
40
- const reachedIdle = await waitForIdle(params.conversationId);
41
- if (!reachedIdle) {
42
- log.warn(
43
- { conversationId: params.conversationId },
44
- "Timed out waiting for conversation idle; injecting anyway",
45
- );
46
- }
47
- }
48
-
49
- const msg = await addMessage(
50
- params.conversationId,
51
- "assistant",
52
- JSON.stringify([{ type: "text", text: params.text }]),
53
- { skipIndexing: true },
54
- );
55
-
56
- const current = findConversation(params.conversationId);
57
- if (current && !current.isProcessing()) {
58
- current.getMessages().push(createAssistantMessage(params.text));
59
-
60
- params.broadcastMessage({
61
- type: "assistant_text_delta",
62
- text: params.text,
63
- conversationId: params.conversationId,
64
- });
65
- params.broadcastMessage({
66
- type: "message_complete",
67
- conversationId: params.conversationId,
68
- messageId: msg.id,
69
- source: "aux",
70
- });
71
- }
72
-
73
- // The injected aux message bumps `lastMessageAt`, which can move the
74
- // row in the paginated sidebar window — that's a shape change. macOS
75
- // refreshes its full list on `conversation_list_invalidated`; web
76
- // refreshes via the paired `sync_changed` `conversationsList` tag.
77
- //
78
- // TODO(electron-cutover): drop the `conversation_list_invalidated`
79
- // emission once macOS migrates to the Electron client and consumes
80
- // `sync_changed` directly. See `runtime/sync/resource-sync-events.ts`
81
- // for the symmetric helper used by route handlers.
82
- params.broadcastMessage(
83
- {
84
- type: "conversation_list_invalidated",
85
- reason: "reordered",
86
- },
87
- undefined,
88
- { targetInterfaceId: "macos" },
89
- );
90
- params.broadcastMessage({
91
- type: "sync_changed",
92
- tags: [
93
- SYNC_TAGS.conversationsList,
94
- conversationMessagesSyncTag(params.conversationId),
95
- ],
96
- });
97
- }
@@ -1,226 +0,0 @@
1
- import { describe, expect, test } from "bun:test";
2
-
3
- import {
4
- type DecisionOutput,
5
- formatTranscript,
6
- parseDecisionOutput,
7
- } from "./decision.js";
8
-
9
- describe("parseDecisionOutput", () => {
10
- test("parses SHOULD_BUILD: yes with all fields correctly", () => {
11
- const text = `SHOULD_BUILD: yes
12
- ARTIFACT_TYPE: app
13
- ARTIFACT_TITLE: Sarah's Marathon Training Pace Calculator
14
- ARTIFACT_DESCRIPTION: An interactive pace calculator that accounts for Sarah's goal of a sub-4-hour marathon, her current 9:30/mile easy pace, and the hilly terrain of the Boston course.`;
15
-
16
- const result = parseDecisionOutput(text);
17
- expect(result).toEqual({
18
- shouldBuild: true,
19
- artifactType: "app",
20
- artifactTitle: "Sarah's Marathon Training Pace Calculator",
21
- artifactDescription:
22
- "An interactive pace calculator that accounts for Sarah's goal of a sub-4-hour marathon, her current 9:30/mile easy pace, and the hilly terrain of the Boston course.",
23
- });
24
- });
25
-
26
- test("parses SHOULD_BUILD: no with skip reason", () => {
27
- const text = `SHOULD_BUILD: no
28
- SKIP_REASON: The conversation is too early and generic — the user has only asked a factual question with no personal context.`;
29
-
30
- const result = parseDecisionOutput(text);
31
- expect(result).toEqual({
32
- shouldBuild: false,
33
- skipReason:
34
- "The conversation is too early and generic — the user has only asked a factual question with no personal context.",
35
- });
36
- });
37
-
38
- test("parses SHOULD_BUILD: no without skip reason defaults to 'no reason given'", () => {
39
- const text = `SHOULD_BUILD: no`;
40
-
41
- const result = parseDecisionOutput(text);
42
- expect(result).toEqual({
43
- shouldBuild: false,
44
- skipReason: "no reason given",
45
- });
46
- });
47
-
48
- test("returns null for missing SHOULD_BUILD line", () => {
49
- const text = `ARTIFACT_TYPE: app
50
- ARTIFACT_TITLE: Some Title
51
- ARTIFACT_DESCRIPTION: Some description.`;
52
-
53
- const result = parseDecisionOutput(text);
54
- expect(result).toBeNull();
55
- });
56
-
57
- test("returns null when yes but ARTIFACT_TYPE is missing", () => {
58
- const text = `SHOULD_BUILD: yes
59
- ARTIFACT_TITLE: Some Title
60
- ARTIFACT_DESCRIPTION: Some description.`;
61
-
62
- const result = parseDecisionOutput(text);
63
- expect(result).toBeNull();
64
- });
65
-
66
- test("returns null when yes but ARTIFACT_TYPE is invalid", () => {
67
- const text = `SHOULD_BUILD: yes
68
- ARTIFACT_TYPE: widget
69
- ARTIFACT_TITLE: Some Title
70
- ARTIFACT_DESCRIPTION: Some description.`;
71
-
72
- const result = parseDecisionOutput(text);
73
- expect(result).toBeNull();
74
- });
75
-
76
- test("returns null when yes but ARTIFACT_TITLE is missing", () => {
77
- const text = `SHOULD_BUILD: yes
78
- ARTIFACT_TYPE: document
79
- ARTIFACT_DESCRIPTION: Some description.`;
80
-
81
- const result = parseDecisionOutput(text);
82
- expect(result).toBeNull();
83
- });
84
-
85
- test("returns null when yes but ARTIFACT_DESCRIPTION is missing", () => {
86
- const text = `SHOULD_BUILD: yes
87
- ARTIFACT_TYPE: document
88
- ARTIFACT_TITLE: Some Title`;
89
-
90
- const result = parseDecisionOutput(text);
91
- expect(result).toBeNull();
92
- });
93
-
94
- test("returns null when yes but ARTIFACT_TITLE is empty", () => {
95
- const text = `SHOULD_BUILD: yes
96
- ARTIFACT_TYPE: app
97
- ARTIFACT_TITLE:
98
- ARTIFACT_DESCRIPTION: Some description.`;
99
-
100
- const result = parseDecisionOutput(text);
101
- expect(result).toBeNull();
102
- });
103
-
104
- test("handles multi-line ARTIFACT_DESCRIPTION", () => {
105
- const text = `SHOULD_BUILD: yes
106
- ARTIFACT_TYPE: document
107
- ARTIFACT_TITLE: Jake's Q3 OKR Tracker
108
- ARTIFACT_DESCRIPTION: A structured comparison table for Jake's three competing priorities:
109
- scaling the data pipeline from 10M to 50M events/day,
110
- hiring two senior engineers by September,
111
- and reducing p99 latency below 200ms.`;
112
-
113
- const result = parseDecisionOutput(text) as DecisionOutput & {
114
- shouldBuild: true;
115
- };
116
- expect(result).not.toBeNull();
117
- expect(result.shouldBuild).toBe(true);
118
- expect(result.artifactType).toBe("document");
119
- expect(result.artifactTitle).toBe("Jake's Q3 OKR Tracker");
120
- expect(result.artifactDescription).toContain(
121
- "A structured comparison table",
122
- );
123
- expect(result.artifactDescription).toContain(
124
- "reducing p99 latency below 200ms",
125
- );
126
- });
127
-
128
- test("handles case-insensitive SHOULD_BUILD value", () => {
129
- const text = `SHOULD_BUILD: Yes
130
- ARTIFACT_TYPE: app
131
- ARTIFACT_TITLE: Test Title
132
- ARTIFACT_DESCRIPTION: Test description.`;
133
-
134
- const result = parseDecisionOutput(text);
135
- expect(result).not.toBeNull();
136
- expect(result!.shouldBuild).toBe(true);
137
- });
138
-
139
- test("handles case-insensitive ARTIFACT_TYPE value", () => {
140
- const text = `SHOULD_BUILD: yes
141
- ARTIFACT_TYPE: Document
142
- ARTIFACT_TITLE: Test Title
143
- ARTIFACT_DESCRIPTION: Test description.`;
144
-
145
- const result = parseDecisionOutput(text);
146
- expect(result).not.toBeNull();
147
- expect((result as { artifactType: string }).artifactType).toBe("document");
148
- });
149
- });
150
-
151
- describe("formatTranscript", () => {
152
- test("formats plain text messages correctly", () => {
153
- const messages = [
154
- { role: "user", content: "Hello, I need help with my project" },
155
- {
156
- role: "assistant",
157
- content: "I'd be happy to help! What kind of project?",
158
- },
159
- {
160
- role: "user",
161
- content: "I'm building a fitness tracker for marathon training",
162
- },
163
- ];
164
-
165
- const result = formatTranscript(messages);
166
- expect(result).toBe(
167
- `[User]: Hello, I need help with my project
168
-
169
- [Assistant]: I'd be happy to help! What kind of project?
170
-
171
- [User]: I'm building a fitness tracker for marathon training`,
172
- );
173
- });
174
-
175
- test("handles JSON content blocks", () => {
176
- const messages = [
177
- {
178
- role: "user",
179
- content: JSON.stringify([{ type: "text", text: "What is 2+2?" }]),
180
- },
181
- {
182
- role: "assistant",
183
- content: JSON.stringify([
184
- { type: "text", text: "The answer is 4." },
185
- { type: "text", text: "Would you like to know more?" },
186
- ]),
187
- },
188
- ];
189
-
190
- const result = formatTranscript(messages);
191
- expect(result).toBe(
192
- `[User]: What is 2+2?
193
-
194
- [Assistant]: The answer is 4.
195
- Would you like to know more?`,
196
- );
197
- });
198
-
199
- test("handles mixed JSON and plain text messages", () => {
200
- const messages = [
201
- {
202
- role: "user",
203
- content: JSON.stringify([
204
- { type: "text", text: "Help me plan my week" },
205
- ]),
206
- },
207
- { role: "assistant", content: "Sure, what do you have coming up?" },
208
- ];
209
-
210
- const result = formatTranscript(messages);
211
- expect(result).toBe(
212
- `[User]: Help me plan my week
213
-
214
- [Assistant]: Sure, what do you have coming up?`,
215
- );
216
- });
217
-
218
- test("handles unknown roles", () => {
219
- const messages = [
220
- { role: "system", content: "You are a helpful assistant" },
221
- ];
222
-
223
- const result = formatTranscript(messages);
224
- expect(result).toBe("[system]: You are a helpful assistant");
225
- });
226
- });