@illuma-ai/agents 1.1.21 → 1.1.23

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 (244) hide show
  1. package/dist/cjs/graphs/Graph.cjs +12 -1
  2. package/dist/cjs/graphs/Graph.cjs.map +1 -1
  3. package/dist/cjs/graphs/MultiAgentGraph.cjs +105 -1
  4. package/dist/cjs/graphs/MultiAgentGraph.cjs.map +1 -1
  5. package/dist/cjs/run.cjs +20 -9
  6. package/dist/cjs/run.cjs.map +1 -1
  7. package/dist/cjs/utils/llm.cjs.map +1 -1
  8. package/dist/esm/graphs/Graph.mjs +12 -1
  9. package/dist/esm/graphs/Graph.mjs.map +1 -1
  10. package/dist/esm/graphs/MultiAgentGraph.mjs +105 -1
  11. package/dist/esm/graphs/MultiAgentGraph.mjs.map +1 -1
  12. package/dist/esm/run.mjs +20 -9
  13. package/dist/esm/run.mjs.map +1 -1
  14. package/dist/esm/utils/llm.mjs.map +1 -1
  15. package/dist/types/graphs/MultiAgentGraph.d.ts +17 -0
  16. package/package.json +1 -1
  17. package/src/graphs/Graph.ts +13 -1
  18. package/src/graphs/MultiAgentGraph.ts +128 -1
  19. package/src/graphs/__tests__/multi-agent-delegate.test.ts +205 -0
  20. package/src/run.ts +20 -11
  21. package/src/scripts/test-bedrock-handoff-autonomous.ts +231 -0
  22. package/src/utils/llm.ts +1 -0
  23. package/src/agents/AgentContext.js +0 -782
  24. package/src/agents/AgentContext.test.js +0 -421
  25. package/src/agents/__tests__/AgentContext.test.js +0 -678
  26. package/src/agents/__tests__/resolveStructuredOutputMode.test.js +0 -117
  27. package/src/common/enum.js +0 -192
  28. package/src/common/index.js +0 -3
  29. package/src/events.js +0 -166
  30. package/src/graphs/Graph.js +0 -1857
  31. package/src/graphs/MultiAgentGraph.js +0 -1092
  32. package/src/graphs/__tests__/structured-output.integration.test.js +0 -624
  33. package/src/graphs/__tests__/structured-output.test.js +0 -144
  34. package/src/graphs/contextManagement.e2e.test.js +0 -718
  35. package/src/graphs/contextManagement.test.js +0 -485
  36. package/src/graphs/handoffValidation.test.js +0 -276
  37. package/src/graphs/index.js +0 -3
  38. package/src/index.js +0 -28
  39. package/src/instrumentation.js +0 -21
  40. package/src/llm/anthropic/index.js +0 -319
  41. package/src/llm/anthropic/types.js +0 -46
  42. package/src/llm/anthropic/utils/message_inputs.js +0 -627
  43. package/src/llm/anthropic/utils/message_outputs.js +0 -290
  44. package/src/llm/anthropic/utils/output_parsers.js +0 -89
  45. package/src/llm/anthropic/utils/tools.js +0 -25
  46. package/src/llm/bedrock/__tests__/bedrock-caching.test.js +0 -392
  47. package/src/llm/bedrock/index.js +0 -303
  48. package/src/llm/bedrock/types.js +0 -2
  49. package/src/llm/bedrock/utils/index.js +0 -6
  50. package/src/llm/bedrock/utils/message_inputs.js +0 -463
  51. package/src/llm/bedrock/utils/message_outputs.js +0 -269
  52. package/src/llm/fake.js +0 -92
  53. package/src/llm/google/index.js +0 -215
  54. package/src/llm/google/types.js +0 -12
  55. package/src/llm/google/utils/common.js +0 -670
  56. package/src/llm/google/utils/tools.js +0 -111
  57. package/src/llm/google/utils/zod_to_genai_parameters.js +0 -47
  58. package/src/llm/openai/index.js +0 -1033
  59. package/src/llm/openai/types.js +0 -2
  60. package/src/llm/openai/utils/index.js +0 -756
  61. package/src/llm/openai/utils/isReasoningModel.test.js +0 -79
  62. package/src/llm/openrouter/index.js +0 -261
  63. package/src/llm/openrouter/reasoning.test.js +0 -181
  64. package/src/llm/providers.js +0 -36
  65. package/src/llm/text.js +0 -65
  66. package/src/llm/vertexai/index.js +0 -402
  67. package/src/messages/__tests__/tools.test.js +0 -392
  68. package/src/messages/cache.js +0 -404
  69. package/src/messages/cache.test.js +0 -1167
  70. package/src/messages/content.js +0 -48
  71. package/src/messages/content.test.js +0 -314
  72. package/src/messages/core.js +0 -359
  73. package/src/messages/ensureThinkingBlock.test.js +0 -997
  74. package/src/messages/format.js +0 -973
  75. package/src/messages/formatAgentMessages.test.js +0 -2278
  76. package/src/messages/formatAgentMessages.tools.test.js +0 -362
  77. package/src/messages/formatMessage.test.js +0 -608
  78. package/src/messages/ids.js +0 -18
  79. package/src/messages/index.js +0 -9
  80. package/src/messages/labelContentByAgent.test.js +0 -725
  81. package/src/messages/prune.js +0 -438
  82. package/src/messages/reducer.js +0 -60
  83. package/src/messages/shiftIndexTokenCountMap.test.js +0 -63
  84. package/src/messages/summarize.js +0 -146
  85. package/src/messages/summarize.test.js +0 -332
  86. package/src/messages/tools.js +0 -90
  87. package/src/mockStream.js +0 -81
  88. package/src/prompts/collab.js +0 -7
  89. package/src/prompts/index.js +0 -3
  90. package/src/prompts/taskmanager.js +0 -58
  91. package/src/run.js +0 -427
  92. package/src/schemas/index.js +0 -3
  93. package/src/schemas/schema-preparation.test.js +0 -370
  94. package/src/schemas/validate.js +0 -314
  95. package/src/schemas/validate.test.js +0 -264
  96. package/src/scripts/abort.js +0 -127
  97. package/src/scripts/ant_web_search.js +0 -130
  98. package/src/scripts/ant_web_search_edge_case.js +0 -133
  99. package/src/scripts/ant_web_search_error_edge_case.js +0 -119
  100. package/src/scripts/args.js +0 -41
  101. package/src/scripts/bedrock-cache-debug.js +0 -186
  102. package/src/scripts/bedrock-content-aggregation-test.js +0 -195
  103. package/src/scripts/bedrock-merge-test.js +0 -80
  104. package/src/scripts/bedrock-parallel-tools-test.js +0 -150
  105. package/src/scripts/caching.js +0 -106
  106. package/src/scripts/cli.js +0 -152
  107. package/src/scripts/cli2.js +0 -119
  108. package/src/scripts/cli3.js +0 -163
  109. package/src/scripts/cli4.js +0 -165
  110. package/src/scripts/cli5.js +0 -165
  111. package/src/scripts/code_exec.js +0 -171
  112. package/src/scripts/code_exec_files.js +0 -180
  113. package/src/scripts/code_exec_multi_session.js +0 -185
  114. package/src/scripts/code_exec_ptc.js +0 -265
  115. package/src/scripts/code_exec_session.js +0 -217
  116. package/src/scripts/code_exec_simple.js +0 -120
  117. package/src/scripts/content.js +0 -111
  118. package/src/scripts/empty_input.js +0 -125
  119. package/src/scripts/handoff-test.js +0 -96
  120. package/src/scripts/image.js +0 -138
  121. package/src/scripts/memory.js +0 -83
  122. package/src/scripts/multi-agent-chain.js +0 -271
  123. package/src/scripts/multi-agent-conditional.js +0 -185
  124. package/src/scripts/multi-agent-document-review-chain.js +0 -171
  125. package/src/scripts/multi-agent-hybrid-flow.js +0 -264
  126. package/src/scripts/multi-agent-parallel-start.js +0 -214
  127. package/src/scripts/multi-agent-parallel.js +0 -346
  128. package/src/scripts/multi-agent-sequence.js +0 -184
  129. package/src/scripts/multi-agent-supervisor.js +0 -324
  130. package/src/scripts/multi-agent-test.js +0 -147
  131. package/src/scripts/parallel-asymmetric-tools-test.js +0 -202
  132. package/src/scripts/parallel-full-metadata-test.js +0 -176
  133. package/src/scripts/parallel-tools-test.js +0 -256
  134. package/src/scripts/programmatic_exec.js +0 -277
  135. package/src/scripts/programmatic_exec_agent.js +0 -168
  136. package/src/scripts/search.js +0 -118
  137. package/src/scripts/sequential-full-metadata-test.js +0 -143
  138. package/src/scripts/simple.js +0 -174
  139. package/src/scripts/single-agent-metadata-test.js +0 -152
  140. package/src/scripts/stream.js +0 -113
  141. package/src/scripts/test-custom-prompt-key.js +0 -132
  142. package/src/scripts/test-handoff-input.js +0 -143
  143. package/src/scripts/test-handoff-preamble.js +0 -227
  144. package/src/scripts/test-handoff-steering.js +0 -353
  145. package/src/scripts/test-multi-agent-list-handoff.js +0 -318
  146. package/src/scripts/test-parallel-agent-labeling.js +0 -253
  147. package/src/scripts/test-parallel-handoffs.js +0 -229
  148. package/src/scripts/test-thinking-handoff-bedrock.js +0 -132
  149. package/src/scripts/test-thinking-handoff.js +0 -132
  150. package/src/scripts/test-thinking-to-thinking-handoff-bedrock.js +0 -140
  151. package/src/scripts/test-tool-before-handoff-role-order.js +0 -223
  152. package/src/scripts/test-tools-before-handoff.js +0 -187
  153. package/src/scripts/test_code_api.js +0 -263
  154. package/src/scripts/thinking-bedrock.js +0 -128
  155. package/src/scripts/thinking-vertexai.js +0 -130
  156. package/src/scripts/thinking.js +0 -134
  157. package/src/scripts/tool_search.js +0 -114
  158. package/src/scripts/tools.js +0 -125
  159. package/src/specs/agent-handoffs-bedrock.integration.test.js +0 -280
  160. package/src/specs/agent-handoffs.test.js +0 -924
  161. package/src/specs/anthropic.simple.test.js +0 -287
  162. package/src/specs/azure.simple.test.js +0 -381
  163. package/src/specs/cache.simple.test.js +0 -282
  164. package/src/specs/custom-event-await.test.js +0 -148
  165. package/src/specs/deepseek.simple.test.js +0 -189
  166. package/src/specs/emergency-prune.test.js +0 -308
  167. package/src/specs/moonshot.simple.test.js +0 -237
  168. package/src/specs/observability.integration.test.js +0 -1337
  169. package/src/specs/openai.simple.test.js +0 -233
  170. package/src/specs/openrouter.simple.test.js +0 -202
  171. package/src/specs/prune.test.js +0 -733
  172. package/src/specs/reasoning.test.js +0 -144
  173. package/src/specs/spec.utils.js +0 -4
  174. package/src/specs/thinking-handoff.test.js +0 -486
  175. package/src/specs/thinking-prune.test.js +0 -600
  176. package/src/specs/token-distribution-edge-case.test.js +0 -246
  177. package/src/specs/token-memoization.test.js +0 -32
  178. package/src/specs/tokens.test.js +0 -49
  179. package/src/specs/tool-error.test.js +0 -139
  180. package/src/splitStream.js +0 -204
  181. package/src/splitStream.test.js +0 -504
  182. package/src/stream.js +0 -650
  183. package/src/stream.test.js +0 -225
  184. package/src/test/mockTools.js +0 -340
  185. package/src/tools/BrowserTools.js +0 -245
  186. package/src/tools/Calculator.js +0 -38
  187. package/src/tools/Calculator.test.js +0 -225
  188. package/src/tools/CodeExecutor.js +0 -233
  189. package/src/tools/ProgrammaticToolCalling.js +0 -602
  190. package/src/tools/StreamingToolCallBuffer.js +0 -179
  191. package/src/tools/ToolNode.js +0 -930
  192. package/src/tools/ToolSearch.js +0 -904
  193. package/src/tools/__tests__/BrowserTools.test.js +0 -306
  194. package/src/tools/__tests__/ProgrammaticToolCalling.integration.test.js +0 -276
  195. package/src/tools/__tests__/ProgrammaticToolCalling.test.js +0 -807
  196. package/src/tools/__tests__/StreamingToolCallBuffer.test.js +0 -175
  197. package/src/tools/__tests__/ToolApproval.test.js +0 -675
  198. package/src/tools/__tests__/ToolNode.recovery.test.js +0 -200
  199. package/src/tools/__tests__/ToolNode.session.test.js +0 -319
  200. package/src/tools/__tests__/ToolSearch.integration.test.js +0 -125
  201. package/src/tools/__tests__/ToolSearch.test.js +0 -812
  202. package/src/tools/__tests__/handlers.test.js +0 -799
  203. package/src/tools/__tests__/truncation-recovery.integration.test.js +0 -362
  204. package/src/tools/handlers.js +0 -306
  205. package/src/tools/schema.js +0 -25
  206. package/src/tools/search/anthropic.js +0 -34
  207. package/src/tools/search/content.js +0 -116
  208. package/src/tools/search/content.test.js +0 -133
  209. package/src/tools/search/firecrawl.js +0 -173
  210. package/src/tools/search/format.js +0 -198
  211. package/src/tools/search/highlights.js +0 -241
  212. package/src/tools/search/index.js +0 -3
  213. package/src/tools/search/jina-reranker.test.js +0 -106
  214. package/src/tools/search/rerankers.js +0 -165
  215. package/src/tools/search/schema.js +0 -102
  216. package/src/tools/search/search.js +0 -561
  217. package/src/tools/search/serper-scraper.js +0 -126
  218. package/src/tools/search/test.js +0 -129
  219. package/src/tools/search/tool.js +0 -453
  220. package/src/tools/search/types.js +0 -2
  221. package/src/tools/search/utils.js +0 -59
  222. package/src/types/graph.js +0 -24
  223. package/src/types/graph.test.js +0 -192
  224. package/src/types/index.js +0 -7
  225. package/src/types/llm.js +0 -2
  226. package/src/types/messages.js +0 -2
  227. package/src/types/run.js +0 -2
  228. package/src/types/stream.js +0 -2
  229. package/src/types/tools.js +0 -2
  230. package/src/utils/contextAnalytics.js +0 -79
  231. package/src/utils/contextAnalytics.test.js +0 -166
  232. package/src/utils/events.js +0 -26
  233. package/src/utils/graph.js +0 -11
  234. package/src/utils/handlers.js +0 -65
  235. package/src/utils/index.js +0 -10
  236. package/src/utils/llm.js +0 -21
  237. package/src/utils/llmConfig.js +0 -205
  238. package/src/utils/logging.js +0 -37
  239. package/src/utils/misc.js +0 -51
  240. package/src/utils/run.js +0 -69
  241. package/src/utils/schema.js +0 -21
  242. package/src/utils/title.js +0 -119
  243. package/src/utils/tokens.js +0 -92
  244. package/src/utils/toonFormat.js +0 -379
@@ -1,146 +0,0 @@
1
- import { HumanMessage } from '@langchain/core/messages';
2
- const MAX_MESSAGE_CHARS = 4000;
3
- const MAX_TRUNCATED_CHARS = 500;
4
- const SIMPLE_MAX_TOKENS = 512;
5
- export const FULL_SUMMARY_TEMPLATE = `Analyze this conversation and produce a structured summary covering each section below. Be precise and preserve actionable details.
6
-
7
- ## 1. Task Overview & User Intent
8
- What the user originally asked for and their high-level goal.
9
-
10
- ## 2. Technical Context
11
- Languages, frameworks, libraries, APIs, and architectural patterns involved.
12
-
13
- ## 3. Files & Code State
14
- All file paths mentioned, their current state, and any pending modifications.
15
-
16
- ## 4. Problem Resolution History
17
- Problems encountered, debugging steps taken, and their outcomes.
18
-
19
- ## 5. Agent Workflow State
20
- {agent_workflow}
21
-
22
- ## 6. Tool Results Summary
23
- Tools called, their inputs, and key outputs (omit verbose raw data).
24
-
25
- ## 7. Progress Tracking
26
- What has been completed, what remains, and any blockers.
27
-
28
- ## 8. Active Working State
29
- Variables, configurations, and runtime state that must be preserved.
30
-
31
- ## 9. Continuation Plan
32
- Immediate next steps and the order in which they should be executed.
33
-
34
- ## 10. Critical Context
35
- Any constraints, warnings, edge cases, or decisions that must not be lost.
36
-
37
- Conversation:
38
- {conversation}
39
-
40
- Structured Summary:`;
41
- export const SIMPLE_SUMMARY_TEMPLATE = 'Summarize this conversation preserving: user\'s original request, key decisions, current task state, and any file paths mentioned.\n\nConversation:\n{conversation}\n\nSummary:';
42
- export function formatMessagesForSummary(messages) {
43
- const parts = [];
44
- for (const message of messages) {
45
- const role = message.getType();
46
- const raw = typeof message.content === 'string'
47
- ? message.content
48
- : JSON.stringify(message.content);
49
- const content = raw.length > MAX_MESSAGE_CHARS
50
- ? raw.slice(0, MAX_MESSAGE_CHARS) + '...'
51
- : raw;
52
- parts.push(`${role}: ${content}`);
53
- }
54
- return parts.join('\n');
55
- }
56
- export function buildFullSummaryPrompt(conversation, config) {
57
- let agentWorkflow = 'N/A (single-agent conversation)';
58
- if (config?.isMultiAgent === true && config.agentWorkflowState != null) {
59
- const { currentAgentId, agentChain, pendingAgents } = config.agentWorkflowState;
60
- agentWorkflow = [
61
- `Current agent: ${currentAgentId}`,
62
- `Agent chain: ${agentChain.join(' -> ')}`,
63
- `Pending agents: ${pendingAgents.length > 0 ? pendingAgents.join(', ') : 'none'}`,
64
- ].join('\n');
65
- }
66
- return FULL_SUMMARY_TEMPLATE
67
- .replace('{agent_workflow}', agentWorkflow)
68
- .replace('{conversation}', conversation);
69
- }
70
- export function buildSimpleSummaryPrompt(conversation) {
71
- return SIMPLE_SUMMARY_TEMPLATE.replace('{conversation}', conversation);
72
- }
73
- export function createEmergencySummary(messages) {
74
- let firstUserContent = '';
75
- let lastAIContent = '';
76
- const toolNames = new Set();
77
- for (const message of messages) {
78
- const type = message.getType();
79
- const content = typeof message.content === 'string'
80
- ? message.content
81
- : JSON.stringify(message.content);
82
- if (type === 'human' && firstUserContent.length === 0) {
83
- firstUserContent = content.slice(0, MAX_TRUNCATED_CHARS);
84
- }
85
- if (type === 'ai') {
86
- lastAIContent = content.slice(0, MAX_TRUNCATED_CHARS);
87
- }
88
- if (type === 'tool') {
89
- const name = message.name;
90
- if (name != null && name.length > 0) {
91
- toolNames.add(name);
92
- }
93
- }
94
- }
95
- const toolList = toolNames.size > 0 ? [...toolNames].join(', ') : 'none';
96
- return [
97
- '[Emergency Context Summary]',
98
- `Original request: ${firstUserContent}`,
99
- `Last response: ${lastAIContent}`,
100
- `Tools used: ${toolList}`,
101
- `Messages compacted: ${messages.length}`,
102
- ].join('\n');
103
- }
104
- export function validateSummarySize(summary, budget, tokenCounter) {
105
- const count = tokenCounter(new HumanMessage(summary));
106
- return count <= budget;
107
- }
108
- export async function summarize(messages, callback, config) {
109
- const messageCount = messages.length;
110
- const maxOutputTokens = config?.maxOutputTokens ?? 1024;
111
- // Tier 1: Full structured summary
112
- try {
113
- const conversation = formatMessagesForSummary(messages);
114
- const prompt = buildFullSummaryPrompt(conversation, config);
115
- const result = await callback(prompt, maxOutputTokens);
116
- if (result != null && result.length > 0) {
117
- if (config?.summaryBudget != null && config.summaryBudget > 0 && config.tokenCounter != null) {
118
- if (validateSummarySize(result, config.summaryBudget, config.tokenCounter)) {
119
- return { summary: result, tier: 'full', messagesCompacted: messageCount };
120
- }
121
- }
122
- else {
123
- return { summary: result, tier: 'full', messagesCompacted: messageCount };
124
- }
125
- }
126
- }
127
- catch {
128
- // Fall through to Tier 2
129
- }
130
- // Tier 2: Simple summary
131
- try {
132
- const conversation = formatMessagesForSummary(messages);
133
- const prompt = buildSimpleSummaryPrompt(conversation);
134
- const result = await callback(prompt, SIMPLE_MAX_TOKENS);
135
- if (result != null && result.length > 0) {
136
- return { summary: result, tier: 'simple', messagesCompacted: messageCount };
137
- }
138
- }
139
- catch {
140
- // Fall through to Tier 3
141
- }
142
- // Tier 3: Emergency (no LLM)
143
- const summary = createEmergencySummary(messages);
144
- return { summary, tier: 'emergency', messagesCompacted: messageCount };
145
- }
146
- //# sourceMappingURL=summarize.js.map
@@ -1,332 +0,0 @@
1
- import { HumanMessage, AIMessage, SystemMessage, ToolMessage, } from '@langchain/core/messages';
2
- import { formatMessagesForSummary, buildFullSummaryPrompt, buildSimpleSummaryPrompt, createEmergencySummary, validateSummarySize, summarize, } from './summarize';
3
- /**
4
- * Simple token counter that approximates 1 token per 4 characters.
5
- * Used across all tests that need a TokenCounter.
6
- */
7
- const simpleTokenCounter = (msg) => {
8
- const content = typeof msg.content === 'string' ? msg.content : JSON.stringify(msg.content);
9
- return Math.ceil(content.length / 4);
10
- };
11
- describe('formatMessagesForSummary', () => {
12
- it('formats messages as "role: content" pairs', () => {
13
- const messages = [
14
- new HumanMessage('Hello there'),
15
- new AIMessage('Hi! How can I help?'),
16
- new SystemMessage('You are a helpful assistant'),
17
- ];
18
- const result = formatMessagesForSummary(messages);
19
- expect(result).toBe('human: Hello there\nai: Hi! How can I help?\nsystem: You are a helpful assistant');
20
- });
21
- it('truncates long content (>4000 chars) with ellipsis', () => {
22
- const longContent = 'x'.repeat(5000);
23
- const messages = [new HumanMessage(longContent)];
24
- const result = formatMessagesForSummary(messages);
25
- // 4000 chars of content + "..." + "human: " prefix
26
- expect(result).toBe(`human: ${'x'.repeat(4000)}...`);
27
- expect(result.length).toBe(4000 + 3 + 7); // content + ellipsis + "human: "
28
- });
29
- it('does not truncate content at exactly 4000 chars', () => {
30
- const exactContent = 'y'.repeat(4000);
31
- const messages = [new HumanMessage(exactContent)];
32
- const result = formatMessagesForSummary(messages);
33
- expect(result).toBe(`human: ${exactContent}`);
34
- expect(result).not.toContain('...');
35
- });
36
- it('handles empty messages array', () => {
37
- const result = formatMessagesForSummary([]);
38
- expect(result).toBe('');
39
- });
40
- it('handles messages with complex (array) content', () => {
41
- const messages = [
42
- new HumanMessage({
43
- content: [
44
- { type: 'text', text: 'Look at this image' },
45
- { type: 'image_url', image_url: { url: 'data:image/png;base64,abc123' } },
46
- ],
47
- }),
48
- ];
49
- const result = formatMessagesForSummary(messages);
50
- // Array content gets JSON.stringified
51
- expect(result).toContain('human: ');
52
- expect(result).toContain('Look at this image');
53
- expect(result).toContain('image_url');
54
- });
55
- });
56
- describe('buildFullSummaryPrompt', () => {
57
- const conversation = 'human: Hello\nai: Hi there!';
58
- it('contains all 10 section headers', () => {
59
- const prompt = buildFullSummaryPrompt(conversation);
60
- const expectedSections = [
61
- '## 1. Task Overview & User Intent',
62
- '## 2. Technical Context',
63
- '## 3. Files & Code State',
64
- '## 4. Problem Resolution History',
65
- '## 5. Agent Workflow State',
66
- '## 6. Tool Results Summary',
67
- '## 7. Progress Tracking',
68
- '## 8. Active Working State',
69
- '## 9. Continuation Plan',
70
- '## 10. Critical Context',
71
- ];
72
- for (const section of expectedSections) {
73
- expect(prompt).toContain(section);
74
- }
75
- });
76
- it('includes conversation text', () => {
77
- const prompt = buildFullSummaryPrompt(conversation);
78
- expect(prompt).toContain(conversation);
79
- });
80
- it('includes agent workflow state when isMultiAgent=true', () => {
81
- const prompt = buildFullSummaryPrompt(conversation, {
82
- isMultiAgent: true,
83
- agentWorkflowState: {
84
- currentAgentId: 'coder',
85
- agentChain: ['planner', 'coder'],
86
- pendingAgents: ['reviewer'],
87
- },
88
- });
89
- expect(prompt).toContain('Current agent: coder');
90
- expect(prompt).toContain('Agent chain: planner -> coder');
91
- expect(prompt).toContain('Pending agents: reviewer');
92
- expect(prompt).not.toContain('N/A (single-agent conversation)');
93
- });
94
- it('shows "none" for empty pendingAgents in multi-agent mode', () => {
95
- const prompt = buildFullSummaryPrompt(conversation, {
96
- isMultiAgent: true,
97
- agentWorkflowState: {
98
- currentAgentId: 'coder',
99
- agentChain: ['coder'],
100
- pendingAgents: [],
101
- },
102
- });
103
- expect(prompt).toContain('Pending agents: none');
104
- });
105
- it('omits agent workflow when not multi-agent', () => {
106
- const prompt = buildFullSummaryPrompt(conversation);
107
- expect(prompt).toContain('N/A (single-agent conversation)');
108
- });
109
- it('omits agent workflow when isMultiAgent is true but no state provided', () => {
110
- const prompt = buildFullSummaryPrompt(conversation, { isMultiAgent: true });
111
- expect(prompt).toContain('N/A (single-agent conversation)');
112
- });
113
- });
114
- describe('buildSimpleSummaryPrompt', () => {
115
- it('contains "Summarize this conversation" directive', () => {
116
- const prompt = buildSimpleSummaryPrompt('human: test');
117
- expect(prompt).toContain('Summarize this conversation');
118
- });
119
- it('includes conversation text', () => {
120
- const conversation = 'human: What is 2+2?\nai: 4';
121
- const prompt = buildSimpleSummaryPrompt(conversation);
122
- expect(prompt).toContain(conversation);
123
- });
124
- });
125
- describe('createEmergencySummary', () => {
126
- it('extracts first user message', () => {
127
- const messages = [
128
- new HumanMessage('First question'),
129
- new HumanMessage('Second question'),
130
- ];
131
- const result = createEmergencySummary(messages);
132
- expect(result).toContain('Original request: First question');
133
- });
134
- it('extracts last AI message', () => {
135
- const messages = [
136
- new AIMessage('First answer'),
137
- new AIMessage('Second answer'),
138
- ];
139
- const result = createEmergencySummary(messages);
140
- // Should contain the LAST AI message, not the first
141
- expect(result).toContain('Last response: Second answer');
142
- });
143
- it('extracts tool names from tool messages', () => {
144
- const messages = [
145
- new ToolMessage({ content: 'result1', tool_call_id: 'tc1', name: 'search' }),
146
- new ToolMessage({ content: 'result2', tool_call_id: 'tc2', name: 'calculator' }),
147
- new ToolMessage({ content: 'result3', tool_call_id: 'tc3', name: 'search' }), // duplicate
148
- ];
149
- const result = createEmergencySummary(messages);
150
- expect(result).toContain('Tools used: search, calculator');
151
- });
152
- it('reports correct message count', () => {
153
- const messages = [
154
- new HumanMessage('q1'),
155
- new AIMessage('a1'),
156
- new HumanMessage('q2'),
157
- new AIMessage('a2'),
158
- new ToolMessage({ content: 'r', tool_call_id: 'tc1', name: 'tool1' }),
159
- ];
160
- const result = createEmergencySummary(messages);
161
- expect(result).toContain('Messages compacted: 5');
162
- });
163
- it('handles empty array', () => {
164
- const result = createEmergencySummary([]);
165
- expect(result).toContain('[Emergency Context Summary]');
166
- expect(result).toContain('Original request: ');
167
- expect(result).toContain('Last response: ');
168
- expect(result).toContain('Tools used: none');
169
- expect(result).toContain('Messages compacted: 0');
170
- });
171
- it('handles messages with no user or AI messages', () => {
172
- const messages = [
173
- new SystemMessage('You are helpful'),
174
- new ToolMessage({ content: 'ok', tool_call_id: 'tc1', name: 'ping' }),
175
- ];
176
- const result = createEmergencySummary(messages);
177
- expect(result).toContain('Original request: ');
178
- expect(result).toContain('Last response: ');
179
- expect(result).toContain('Tools used: ping');
180
- expect(result).toContain('Messages compacted: 2');
181
- });
182
- it('truncates long content to 500 chars', () => {
183
- const longContent = 'a'.repeat(1000);
184
- const messages = [new HumanMessage(longContent)];
185
- const result = createEmergencySummary(messages);
186
- // The original request should be truncated to 500 chars
187
- expect(result).toContain(`Original request: ${'a'.repeat(500)}`);
188
- expect(result).not.toContain('a'.repeat(501));
189
- });
190
- });
191
- describe('validateSummarySize', () => {
192
- it('returns true when summary fits in budget', () => {
193
- const summary = 'Short summary';
194
- // simpleTokenCounter: ceil(13/4) = 4 tokens
195
- const result = validateSummarySize(summary, 100, simpleTokenCounter);
196
- expect(result).toBe(true);
197
- });
198
- it('returns false when summary exceeds budget', () => {
199
- const summary = 'This is a summary that should exceed a tiny budget';
200
- // simpleTokenCounter: ceil(50/4) = 13 tokens
201
- const result = validateSummarySize(summary, 1, simpleTokenCounter);
202
- expect(result).toBe(false);
203
- });
204
- it('uses the provided tokenCounter', () => {
205
- const customCounter = jest.fn().mockReturnValue(50);
206
- validateSummarySize('any text', 100, customCounter);
207
- expect(customCounter).toHaveBeenCalledTimes(1);
208
- // Should be called with a HumanMessage wrapping the summary text
209
- const arg = customCounter.mock.calls[0][0];
210
- expect(arg).toBeInstanceOf(HumanMessage);
211
- expect(arg.content).toBe('any text');
212
- });
213
- });
214
- describe('summarize', () => {
215
- const sampleMessages = [
216
- new HumanMessage('What is TypeScript?'),
217
- new AIMessage('TypeScript is a typed superset of JavaScript.'),
218
- new ToolMessage({ content: 'tool result', tool_call_id: 'tc1', name: 'search' }),
219
- ];
220
- it('Tier 1 success: callback returns valid summary -> result.tier === "full"', async () => {
221
- const mockCallback = jest.fn().mockResolvedValueOnce('## Summary\nThis is a valid summary');
222
- const result = await summarize(sampleMessages, mockCallback);
223
- expect(result.tier).toBe('full');
224
- expect(result.summary).toBe('## Summary\nThis is a valid summary');
225
- expect(mockCallback).toHaveBeenCalledTimes(1);
226
- });
227
- it('Tier 1 fails -> Tier 2 success: first callback throws, second succeeds', async () => {
228
- const mockCallback = jest.fn()
229
- .mockRejectedValueOnce(new Error('LLM overloaded'))
230
- .mockResolvedValueOnce('Simple summary');
231
- const result = await summarize(sampleMessages, mockCallback);
232
- expect(result.tier).toBe('simple');
233
- expect(result.summary).toBe('Simple summary');
234
- expect(mockCallback).toHaveBeenCalledTimes(2);
235
- });
236
- it('Tier 1 + 2 fail -> Tier 3: callback always throws', async () => {
237
- const mockCallback = jest.fn().mockRejectedValue(new Error('LLM down'));
238
- const result = await summarize(sampleMessages, mockCallback);
239
- expect(result.tier).toBe('emergency');
240
- expect(result.summary).toContain('[Emergency Context Summary]');
241
- expect(result.summary).toContain('What is TypeScript?');
242
- expect(mockCallback).toHaveBeenCalledTimes(2);
243
- });
244
- it('Callback returns undefined -> falls to Tier 2', async () => {
245
- const mockCallback = jest.fn()
246
- .mockResolvedValueOnce(undefined)
247
- .mockResolvedValueOnce('Tier 2 summary');
248
- const result = await summarize(sampleMessages, mockCallback);
249
- expect(result.tier).toBe('simple');
250
- expect(result.summary).toBe('Tier 2 summary');
251
- expect(mockCallback).toHaveBeenCalledTimes(2);
252
- });
253
- it('Callback returns empty string -> falls to Tier 2', async () => {
254
- const mockCallback = jest.fn()
255
- .mockResolvedValueOnce('')
256
- .mockResolvedValueOnce('Tier 2 result');
257
- const result = await summarize(sampleMessages, mockCallback);
258
- expect(result.tier).toBe('simple');
259
- expect(result.summary).toBe('Tier 2 result');
260
- });
261
- it('Empty messages: returns emergency summary immediately', async () => {
262
- const mockCallback = jest.fn().mockResolvedValue('should not matter');
263
- const result = await summarize([], mockCallback);
264
- // With empty messages, tier 1 still runs (formatMessagesForSummary returns ''),
265
- // but callback returns content so it should succeed as tier 1
266
- // unless the callback result is validated against budget.
267
- // Actually, empty messages still go through the normal flow.
268
- // The callback returns a valid string, so it should be 'full'.
269
- expect(result.messagesCompacted).toBe(0);
270
- });
271
- it('Token budget validation: summary exceeds budget -> falls to next tier', async () => {
272
- const longSummary = 'x'.repeat(2000);
273
- const mockCallback = jest.fn()
274
- .mockResolvedValueOnce(longSummary) // Tier 1: exceeds budget
275
- .mockResolvedValueOnce('Short tier 2 summary'); // Tier 2: no budget check
276
- const result = await summarize(sampleMessages, mockCallback, {
277
- summaryBudget: 10, // Very small budget: 10 tokens
278
- tokenCounter: simpleTokenCounter,
279
- });
280
- expect(result.tier).toBe('simple');
281
- expect(result.summary).toBe('Short tier 2 summary');
282
- expect(mockCallback).toHaveBeenCalledTimes(2);
283
- });
284
- it('Token budget validation: summary fits within budget -> stays at Tier 1', async () => {
285
- const mockCallback = jest.fn().mockResolvedValueOnce('Short summary');
286
- const result = await summarize(sampleMessages, mockCallback, {
287
- summaryBudget: 1000,
288
- tokenCounter: simpleTokenCounter,
289
- });
290
- expect(result.tier).toBe('full');
291
- expect(result.summary).toBe('Short summary');
292
- expect(mockCallback).toHaveBeenCalledTimes(1);
293
- });
294
- it('messagesCompacted count matches input length', async () => {
295
- const mockCallback = jest.fn().mockResolvedValueOnce('Valid summary');
296
- const messages = [
297
- new HumanMessage('q1'),
298
- new AIMessage('a1'),
299
- new HumanMessage('q2'),
300
- new AIMessage('a2'),
301
- new HumanMessage('q3'),
302
- ];
303
- const result = await summarize(messages, mockCallback);
304
- expect(result.messagesCompacted).toBe(5);
305
- });
306
- it('messagesCompacted count is correct even for emergency tier', async () => {
307
- const mockCallback = jest.fn().mockRejectedValue(new Error('fail'));
308
- const result = await summarize(sampleMessages, mockCallback);
309
- expect(result.tier).toBe('emergency');
310
- expect(result.messagesCompacted).toBe(3);
311
- });
312
- it('passes maxOutputTokens to Tier 1 callback', async () => {
313
- const mockCallback = jest.fn().mockResolvedValueOnce('Summary');
314
- await summarize(sampleMessages, mockCallback, { maxOutputTokens: 2048 });
315
- expect(mockCallback).toHaveBeenCalledWith(expect.any(String), 2048);
316
- });
317
- it('defaults maxOutputTokens to 1024 for Tier 1', async () => {
318
- const mockCallback = jest.fn().mockResolvedValueOnce('Summary');
319
- await summarize(sampleMessages, mockCallback);
320
- expect(mockCallback).toHaveBeenCalledWith(expect.any(String), 1024);
321
- });
322
- it('passes 512 as maxTokens for Tier 2 (simple) callback', async () => {
323
- const mockCallback = jest.fn()
324
- .mockRejectedValueOnce(new Error('Tier 1 fail'))
325
- .mockResolvedValueOnce('Simple summary');
326
- await summarize(sampleMessages, mockCallback);
327
- // Second call (Tier 2) should use SIMPLE_MAX_TOKENS = 512
328
- expect(mockCallback).toHaveBeenCalledTimes(2);
329
- expect(mockCallback.mock.calls[1][1]).toBe(512);
330
- });
331
- });
332
- //# sourceMappingURL=summarize.test.js.map
@@ -1,90 +0,0 @@
1
- // src/messages/toolDiscovery.ts
2
- import { AIMessageChunk } from '@langchain/core/messages';
3
- import { Constants, MessageTypes } from '@/common';
4
- import { findLastIndex } from './core';
5
- /**
6
- * Extracts discovered tool names from tool search results in the current turn.
7
- * Only processes tool search messages after the latest AI message with tool calls.
8
- *
9
- * Similar pattern to formatArtifactPayload - finds relevant messages efficiently
10
- * by identifying the latest AI parent and only processing subsequent tool messages.
11
- *
12
- * @param messages - All messages in the conversation
13
- * @returns Array of discovered tool names (empty if no new discoveries)
14
- */
15
- export function extractToolDiscoveries(messages) {
16
- if (messages.length === 0)
17
- return [];
18
- const lastMessage = messages[messages.length - 1];
19
- // Use getType() instead of instanceof to avoid module mismatch issues
20
- if (lastMessage.getType() !== MessageTypes.TOOL)
21
- return [];
22
- const lastToolMessage = lastMessage;
23
- // Find the latest AIMessage with tool_calls that this tool message belongs to
24
- const latestAIParentIndex = findLastIndex(messages, (msg) => (msg instanceof AIMessageChunk &&
25
- (msg.tool_calls?.length ?? 0) > 0 &&
26
- msg.tool_calls?.some((tc) => tc.id === lastToolMessage.tool_call_id)) ??
27
- false);
28
- if (latestAIParentIndex === -1)
29
- return [];
30
- // Collect tool_call_ids from the AI message
31
- const aiMessage = messages[latestAIParentIndex];
32
- const toolCallIds = new Set(aiMessage.tool_calls?.map((tc) => tc.id) ?? []);
33
- // Only process tool search results after the AI message that belong to this turn
34
- const discoveredNames = [];
35
- for (let i = latestAIParentIndex + 1; i < messages.length; i++) {
36
- const msg = messages[i];
37
- // Use getType() instead of instanceof to avoid module mismatch issues
38
- if (msg.getType() !== MessageTypes.TOOL)
39
- continue;
40
- const toolMsg = msg;
41
- if (toolMsg.name !== Constants.TOOL_SEARCH)
42
- continue;
43
- if (!toolCallIds.has(toolMsg.tool_call_id))
44
- continue;
45
- // This is a tool search result from the current turn
46
- if (typeof toolMsg.artifact === 'object' && toolMsg.artifact != null) {
47
- const artifact = toolMsg.artifact;
48
- if (artifact.tool_references && artifact.tool_references.length > 0) {
49
- for (const ref of artifact.tool_references) {
50
- discoveredNames.push(ref.tool_name);
51
- }
52
- }
53
- }
54
- }
55
- return discoveredNames;
56
- }
57
- /**
58
- * Checks if the current turn has any tool search results.
59
- * Quick check to avoid full extraction when not needed.
60
- */
61
- export function hasToolSearchInCurrentTurn(messages) {
62
- if (messages.length === 0)
63
- return false;
64
- const lastMessage = messages[messages.length - 1];
65
- // Use getType() instead of instanceof to avoid module mismatch issues
66
- if (lastMessage.getType() !== MessageTypes.TOOL)
67
- return false;
68
- const lastToolMessage = lastMessage;
69
- // Find the latest AIMessage with tool_calls
70
- const latestAIParentIndex = findLastIndex(messages, (msg) => (msg instanceof AIMessageChunk &&
71
- (msg.tool_calls?.length ?? 0) > 0 &&
72
- msg.tool_calls?.some((tc) => tc.id === lastToolMessage.tool_call_id)) ??
73
- false);
74
- if (latestAIParentIndex === -1)
75
- return false;
76
- const aiMessage = messages[latestAIParentIndex];
77
- const toolCallIds = new Set(aiMessage.tool_calls?.map((tc) => tc.id) ?? []);
78
- // Check if any tool search results exist after the AI message
79
- // Use getType() instead of instanceof to avoid module mismatch issues
80
- for (let i = latestAIParentIndex + 1; i < messages.length; i++) {
81
- const msg = messages[i];
82
- if (msg.getType() === MessageTypes.TOOL &&
83
- msg.name === Constants.TOOL_SEARCH &&
84
- toolCallIds.has(msg.tool_call_id)) {
85
- return true;
86
- }
87
- }
88
- return false;
89
- }
90
- //# sourceMappingURL=tools.js.map
package/src/mockStream.js DELETED
@@ -1,81 +0,0 @@
1
- // import { SplitStreamHandler } from '@/splitStream';
2
- // import { GraphEvents } from '@/common';
3
- import { sleep } from '@/utils';
4
- const choiceProps = { finish_reason: null, index: 0, delta: {} };
5
- const reasoningSplitRegex = /(?<=\s+)|(?=\s+)/;
6
- const contentSplitRegex = /(?<=<\/?think>)|(?=<\/?think>)|(?<=\s+)|(?=\s+)/;
7
- export const createMockStream = (options = {}) => {
8
- const { text, reasoningText, streamRate = 25, reasoningKey = 'reasoning_content' } = options;
9
- return async function* mockOpenAIStream() {
10
- const content = text ?? `Here's a sample message that includes code:
11
- \`\`\`python
12
- def hello_world():
13
- print("Hello, World!")
14
- # This is a long code block
15
- # That shouldn't be split
16
- return True
17
- \`\`\`
18
- Now we're back to regular text. This is a very long sentence that should probably be split at some point because it exceeds our threshold and contains multiple natural breaking points. Let's see how it handles this case properly.
19
-
20
- Here's another code block:
21
- \`\`\`javascript
22
- console.log("Another test");
23
- // More code here
24
- \`\`\`
25
- And finally some more regular text to test our splitting logic.`;
26
- if (reasoningText != null && reasoningText) {
27
- // Split reasoning text into "token-like" chunks
28
- const reasoningTokens = reasoningText.split(reasoningSplitRegex);
29
- for (const token of reasoningTokens) {
30
- yield {
31
- choices: [{
32
- ...choiceProps,
33
- delta: {
34
- [reasoningKey]: token,
35
- },
36
- }]
37
- };
38
- await sleep(streamRate);
39
- }
40
- }
41
- // Split main content into "token-like" chunks
42
- const tokens = content.split(contentSplitRegex);
43
- for (const token of tokens) {
44
- yield {
45
- choices: [{
46
- ...choiceProps,
47
- delta: {
48
- content: token
49
- }
50
- }]
51
- };
52
- await sleep(streamRate);
53
- }
54
- };
55
- };
56
- /**
57
- (async function testStream(): Promise<void> {
58
- const runId = nanoid();
59
-
60
- const streamHandler = new SplitStreamHandler({
61
- runId,
62
- handlers: {
63
- [GraphEvents.ON_RUN_STEP]: (data): void => {
64
- console.dir(data, { depth: null });
65
- },
66
- [GraphEvents.ON_MESSAGE_DELTA]: (): void => {
67
- // console.dir(data, { depth: null });
68
- },
69
- },
70
- });
71
- const stream = createMockStream({
72
- reasoningText: 'This is a test reasoning text.',
73
- streamRate: 5,
74
- })();
75
-
76
- for await (const chunk of stream) {
77
- streamHandler.handle(chunk);
78
- }
79
- })();
80
- */
81
- //# sourceMappingURL=mockStream.js.map
@@ -1,7 +0,0 @@
1
- // src/prompts/collab.ts
2
- export const supervisorPrompt = `You are a supervisor tasked with managing a conversation between the
3
- following workers: {members}. Given the following user request,
4
- respond with the worker to act next. Each worker will perform a
5
- task and respond with their results and status. Multiple workers can work at once, and they can use multiple tools at once. Each worker can run their tools multiple times per task. When finished,
6
- respond with FINISH.`;
7
- //# sourceMappingURL=collab.js.map
@@ -1,3 +0,0 @@
1
- export * from './collab';
2
- export * from './taskmanager';
3
- //# sourceMappingURL=index.js.map