darwin-agents 0.7.1 → 0.9.0

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 (201) hide show
  1. package/CHANGELOG.md +62 -0
  2. package/README.md +62 -1
  3. package/dist/src/cli/canary.d.ts +13 -0
  4. package/dist/src/cli/canary.d.ts.map +1 -0
  5. package/dist/src/cli/canary.js +66 -0
  6. package/dist/src/cli/canary.js.map +1 -0
  7. package/dist/src/cli/evolution-flags.d.ts +40 -0
  8. package/dist/src/cli/evolution-flags.d.ts.map +1 -0
  9. package/dist/src/cli/evolution-flags.js +106 -0
  10. package/dist/src/cli/evolution-flags.js.map +1 -0
  11. package/dist/src/cli/evolve.d.ts +10 -1
  12. package/dist/src/cli/evolve.d.ts.map +1 -1
  13. package/dist/src/cli/evolve.js +93 -8
  14. package/dist/src/cli/evolve.js.map +1 -1
  15. package/dist/src/cli/index.js +18 -0
  16. package/dist/src/cli/index.js.map +1 -1
  17. package/dist/src/cli/run.d.ts.map +1 -1
  18. package/dist/src/cli/run.js +54 -70
  19. package/dist/src/cli/run.js.map +1 -1
  20. package/dist/src/evolution/build-loop.d.ts +20 -0
  21. package/dist/src/evolution/build-loop.d.ts.map +1 -0
  22. package/dist/src/evolution/build-loop.js +67 -0
  23. package/dist/src/evolution/build-loop.js.map +1 -0
  24. package/dist/src/evolution/canary.d.ts +164 -0
  25. package/dist/src/evolution/canary.d.ts.map +1 -0
  26. package/dist/src/evolution/canary.js +244 -0
  27. package/dist/src/evolution/canary.js.map +1 -0
  28. package/dist/src/evolution/critic-families.d.ts +51 -0
  29. package/dist/src/evolution/critic-families.d.ts.map +1 -0
  30. package/dist/src/evolution/critic-families.js +51 -0
  31. package/dist/src/evolution/critic-families.js.map +1 -0
  32. package/dist/src/evolution/enabled-state.d.ts +59 -0
  33. package/dist/src/evolution/enabled-state.d.ts.map +1 -0
  34. package/dist/src/evolution/enabled-state.js +122 -0
  35. package/dist/src/evolution/enabled-state.js.map +1 -0
  36. package/dist/src/evolution/loop.d.ts +46 -0
  37. package/dist/src/evolution/loop.d.ts.map +1 -1
  38. package/dist/src/evolution/loop.js +119 -1
  39. package/dist/src/evolution/loop.js.map +1 -1
  40. package/dist/src/evolution/parse-score.d.ts +35 -0
  41. package/dist/src/evolution/parse-score.d.ts.map +1 -0
  42. package/dist/src/evolution/parse-score.js +115 -0
  43. package/dist/src/evolution/parse-score.js.map +1 -0
  44. package/dist/src/evolution/tracker.d.ts +13 -0
  45. package/dist/src/evolution/tracker.d.ts.map +1 -1
  46. package/dist/src/evolution/tracker.js +34 -0
  47. package/dist/src/evolution/tracker.js.map +1 -1
  48. package/dist/src/index.d.ts +3 -1
  49. package/dist/src/index.d.ts.map +1 -1
  50. package/dist/src/index.js +10 -0
  51. package/dist/src/index.js.map +1 -1
  52. package/dist/src/memory/postgres-memory.d.ts.map +1 -1
  53. package/dist/src/memory/postgres-memory.js +1 -0
  54. package/dist/src/memory/postgres-memory.js.map +1 -1
  55. package/dist/src/memory/sqlite-memory.d.ts.map +1 -1
  56. package/dist/src/memory/sqlite-memory.js +1 -0
  57. package/dist/src/memory/sqlite-memory.js.map +1 -1
  58. package/dist/src/types.d.ts +45 -0
  59. package/dist/src/types.d.ts.map +1 -1
  60. package/dist/src/types.js.map +1 -1
  61. package/package.json +1 -1
  62. package/dist/agents/analyst.d.ts +0 -11
  63. package/dist/agents/analyst.d.ts.map +0 -1
  64. package/dist/agents/analyst.js +0 -78
  65. package/dist/agents/analyst.js.map +0 -1
  66. package/dist/agents/blog-writer.d.ts +0 -13
  67. package/dist/agents/blog-writer.d.ts.map +0 -1
  68. package/dist/agents/blog-writer.js +0 -59
  69. package/dist/agents/blog-writer.js.map +0 -1
  70. package/dist/agents/critic.d.ts +0 -11
  71. package/dist/agents/critic.d.ts.map +0 -1
  72. package/dist/agents/critic.js +0 -57
  73. package/dist/agents/critic.js.map +0 -1
  74. package/dist/agents/index.d.ts +0 -15
  75. package/dist/agents/index.d.ts.map +0 -1
  76. package/dist/agents/index.js +0 -31
  77. package/dist/agents/index.js.map +0 -1
  78. package/dist/agents/investigator-critic.d.ts +0 -10
  79. package/dist/agents/investigator-critic.d.ts.map +0 -1
  80. package/dist/agents/investigator-critic.js +0 -78
  81. package/dist/agents/investigator-critic.js.map +0 -1
  82. package/dist/agents/investigator.d.ts +0 -13
  83. package/dist/agents/investigator.d.ts.map +0 -1
  84. package/dist/agents/investigator.js +0 -105
  85. package/dist/agents/investigator.js.map +0 -1
  86. package/dist/agents/marketing.d.ts +0 -13
  87. package/dist/agents/marketing.d.ts.map +0 -1
  88. package/dist/agents/marketing.js +0 -59
  89. package/dist/agents/marketing.js.map +0 -1
  90. package/dist/agents/researcher.d.ts +0 -11
  91. package/dist/agents/researcher.d.ts.map +0 -1
  92. package/dist/agents/researcher.js +0 -68
  93. package/dist/agents/researcher.js.map +0 -1
  94. package/dist/agents/writer.d.ts +0 -9
  95. package/dist/agents/writer.d.ts.map +0 -1
  96. package/dist/agents/writer.js +0 -47
  97. package/dist/agents/writer.js.map +0 -1
  98. package/dist/cli/create.d.ts +0 -11
  99. package/dist/cli/create.d.ts.map +0 -1
  100. package/dist/cli/create.js +0 -104
  101. package/dist/cli/create.js.map +0 -1
  102. package/dist/cli/evolve.d.ts +0 -13
  103. package/dist/cli/evolve.d.ts.map +0 -1
  104. package/dist/cli/evolve.js +0 -69
  105. package/dist/cli/evolve.js.map +0 -1
  106. package/dist/cli/index.d.ts +0 -13
  107. package/dist/cli/index.d.ts.map +0 -1
  108. package/dist/cli/index.js +0 -84
  109. package/dist/cli/index.js.map +0 -1
  110. package/dist/cli/init.d.ts +0 -12
  111. package/dist/cli/init.d.ts.map +0 -1
  112. package/dist/cli/init.js +0 -68
  113. package/dist/cli/init.js.map +0 -1
  114. package/dist/cli/run.d.ts +0 -7
  115. package/dist/cli/run.d.ts.map +0 -1
  116. package/dist/cli/run.js +0 -371
  117. package/dist/cli/run.js.map +0 -1
  118. package/dist/cli/status.d.ts +0 -7
  119. package/dist/cli/status.d.ts.map +0 -1
  120. package/dist/cli/status.js +0 -123
  121. package/dist/cli/status.js.map +0 -1
  122. package/dist/core/agent.d.ts +0 -53
  123. package/dist/core/agent.d.ts.map +0 -1
  124. package/dist/core/agent.js +0 -172
  125. package/dist/core/agent.js.map +0 -1
  126. package/dist/core/runner.d.ts +0 -75
  127. package/dist/core/runner.d.ts.map +0 -1
  128. package/dist/core/runner.js +0 -255
  129. package/dist/core/runner.js.map +0 -1
  130. package/dist/evolution/loop.d.ts +0 -100
  131. package/dist/evolution/loop.d.ts.map +0 -1
  132. package/dist/evolution/loop.js +0 -424
  133. package/dist/evolution/loop.js.map +0 -1
  134. package/dist/evolution/multi-critic.d.ts +0 -58
  135. package/dist/evolution/multi-critic.d.ts.map +0 -1
  136. package/dist/evolution/multi-critic.js +0 -326
  137. package/dist/evolution/multi-critic.js.map +0 -1
  138. package/dist/evolution/notifications.d.ts +0 -32
  139. package/dist/evolution/notifications.d.ts.map +0 -1
  140. package/dist/evolution/notifications.js +0 -92
  141. package/dist/evolution/notifications.js.map +0 -1
  142. package/dist/evolution/optimizer.d.ts +0 -64
  143. package/dist/evolution/optimizer.d.ts.map +0 -1
  144. package/dist/evolution/optimizer.js +0 -223
  145. package/dist/evolution/optimizer.js.map +0 -1
  146. package/dist/evolution/patterns.d.ts +0 -63
  147. package/dist/evolution/patterns.d.ts.map +0 -1
  148. package/dist/evolution/patterns.js +0 -297
  149. package/dist/evolution/patterns.js.map +0 -1
  150. package/dist/evolution/safety.d.ts +0 -76
  151. package/dist/evolution/safety.d.ts.map +0 -1
  152. package/dist/evolution/safety.js +0 -182
  153. package/dist/evolution/safety.js.map +0 -1
  154. package/dist/evolution/tracker.d.ts +0 -48
  155. package/dist/evolution/tracker.d.ts.map +0 -1
  156. package/dist/evolution/tracker.js +0 -163
  157. package/dist/evolution/tracker.js.map +0 -1
  158. package/dist/index.d.ts +0 -32
  159. package/dist/index.d.ts.map +0 -1
  160. package/dist/index.js +0 -35
  161. package/dist/index.js.map +0 -1
  162. package/dist/memory/index.d.ts +0 -32
  163. package/dist/memory/index.d.ts.map +0 -1
  164. package/dist/memory/index.js +0 -49
  165. package/dist/memory/index.js.map +0 -1
  166. package/dist/memory/postgres-memory.d.ts +0 -52
  167. package/dist/memory/postgres-memory.d.ts.map +0 -1
  168. package/dist/memory/postgres-memory.js +0 -515
  169. package/dist/memory/postgres-memory.js.map +0 -1
  170. package/dist/memory/sqlite-memory.d.ts +0 -36
  171. package/dist/memory/sqlite-memory.d.ts.map +0 -1
  172. package/dist/memory/sqlite-memory.js +0 -380
  173. package/dist/memory/sqlite-memory.js.map +0 -1
  174. package/dist/providers/anthropic.d.ts +0 -20
  175. package/dist/providers/anthropic.d.ts.map +0 -1
  176. package/dist/providers/anthropic.js +0 -82
  177. package/dist/providers/anthropic.js.map +0 -1
  178. package/dist/providers/claude-cli.d.ts +0 -35
  179. package/dist/providers/claude-cli.d.ts.map +0 -1
  180. package/dist/providers/claude-cli.js +0 -153
  181. package/dist/providers/claude-cli.js.map +0 -1
  182. package/dist/providers/index.d.ts +0 -39
  183. package/dist/providers/index.d.ts.map +0 -1
  184. package/dist/providers/index.js +0 -58
  185. package/dist/providers/index.js.map +0 -1
  186. package/dist/providers/ollama.d.ts +0 -17
  187. package/dist/providers/ollama.d.ts.map +0 -1
  188. package/dist/providers/ollama.js +0 -64
  189. package/dist/providers/ollama.js.map +0 -1
  190. package/dist/providers/openai.d.ts +0 -19
  191. package/dist/providers/openai.d.ts.map +0 -1
  192. package/dist/providers/openai.js +0 -75
  193. package/dist/providers/openai.js.map +0 -1
  194. package/dist/providers/types.d.ts +0 -62
  195. package/dist/providers/types.d.ts.map +0 -1
  196. package/dist/providers/types.js +0 -9
  197. package/dist/providers/types.js.map +0 -1
  198. package/dist/types.d.ts +0 -221
  199. package/dist/types.d.ts.map +0 -1
  200. package/dist/types.js +0 -19
  201. package/dist/types.js.map +0 -1
@@ -1,255 +0,0 @@
1
- /**
2
- * Darwin — Agent Runner
3
- *
4
- * Executes agents via pluggable LLM providers.
5
- * Supports: Claude CLI (with MCP), Anthropic API, OpenAI, Ollama.
6
- *
7
- * The runner handles metrics, experiment records, and reports.
8
- * The provider handles the actual LLM call.
9
- */
10
- import { mkdirSync, writeFileSync } from 'node:fs';
11
- import { join, resolve } from 'node:path';
12
- import { ClaudeCliProvider } from '../providers/claude-cli.js';
13
- import { createProvider } from '../providers/index.js';
14
- // ─── Constants ───────────────────────────────────────
15
- const DEFAULT_TIMEOUT = 600_000; // 10 minutes
16
- const DEFAULT_MODEL = 'claude-sonnet-4-20250514';
17
- const DEFAULT_MAX_TURNS = 10;
18
- const DEFAULT_DATA_DIR = '.darwin';
19
- const DEFAULT_TASK_TYPE = 'general';
20
- // ─── Process-lifetime budget caps ────────────────────
21
- //
22
- // Self-evolving agent loops (A/B tests, multi-model critics, convergence
23
- // bugs) can fan out into hundreds of provider calls if nothing stops them,
24
- // and each call hits the billed API. These caps short-circuit the runner
25
- // before the next LLM call so a runaway loop tops out at a known ceiling.
26
- // Both are opt-out (`0` = disabled), overridable via env or the explicit
27
- // setters below. Original feature landed in 29c4367; Round-4 added the
28
- // DarwinBudgetError class + test helpers so regressions can be asserted.
29
- const DEFAULT_MAX_RUNS_PER_PROCESS = 100;
30
- const DEFAULT_MAX_RUN_WALL_MS = 60 * 60 * 1000; // 1 h
31
- let maxRunsPerProcess = Number(process.env.DARWIN_MAX_RUNS_PER_PROCESS ?? DEFAULT_MAX_RUNS_PER_PROCESS);
32
- let maxRunWallMs = Number(process.env.DARWIN_MAX_RUN_WALL_MS ?? DEFAULT_MAX_RUN_WALL_MS);
33
- let runsThisProcess = 0;
34
- let processStartMs = Date.now();
35
- /** Override the per-process run cap (0 = disabled). Primarily for tests. */
36
- export function setMaxRunsPerProcess(n) {
37
- maxRunsPerProcess = n;
38
- }
39
- /** Override the wall-clock cap in ms (0 = disabled). Primarily for tests. */
40
- export function setMaxRunWallMs(ms) {
41
- maxRunWallMs = ms;
42
- }
43
- /** Reset the process run counter + wall-clock start. Primarily for tests. */
44
- export function resetRunCounters() {
45
- runsThisProcess = 0;
46
- processStartMs = Date.now();
47
- }
48
- /** Error thrown when a process-lifetime budget cap is exceeded. */
49
- export class DarwinBudgetError extends Error {
50
- budget;
51
- constructor(message, budget) {
52
- super(message);
53
- this.budget = budget;
54
- this.name = 'DarwinBudgetError';
55
- }
56
- }
57
- function enforceBudget() {
58
- if (maxRunsPerProcess > 0 && runsThisProcess >= maxRunsPerProcess) {
59
- throw new DarwinBudgetError(`Darwin run cap reached: ${String(runsThisProcess)}/${String(maxRunsPerProcess)} runs in this process. ` +
60
- `Set DARWIN_MAX_RUNS_PER_PROCESS=0 to disable or raise the limit.`, 'runs');
61
- }
62
- const elapsed = Date.now() - processStartMs;
63
- if (maxRunWallMs > 0 && elapsed >= maxRunWallMs) {
64
- throw new DarwinBudgetError(`Darwin wall-clock cap reached: ${String(Math.round(elapsed / 1000))}s / ${String(Math.round(maxRunWallMs / 1000))}s. ` +
65
- `Restart the process, or set DARWIN_MAX_RUN_WALL_MS=0 to disable.`, 'wall');
66
- }
67
- }
68
- // Matches common URL patterns for source counting
69
- const URL_PATTERN = /https?:\/\/[^\s)>\]"'`]+/g;
70
- // ─── Metrics Parser ──────────────────────────────────
71
- /**
72
- * Parse basic metrics from agent output.
73
- * - sourceCount: number of unique URLs found in output
74
- * - outputLength: character count of the output
75
- */
76
- function parseMetrics(output, durationMs) {
77
- const urls = output.match(URL_PATTERN);
78
- const uniqueUrls = urls ? new Set(urls) : new Set();
79
- return {
80
- qualityScore: null, // Set later by evaluator
81
- sourceCount: uniqueUrls.size,
82
- outputLength: output.length,
83
- errorCount: 0,
84
- durationMs,
85
- };
86
- }
87
- // ─── Experiment Factory ──────────────────────────────
88
- function generateExperimentId(agentName) {
89
- const date = new Date().toISOString().slice(0, 10);
90
- const random = Math.random().toString(36).slice(2, 10);
91
- return `exp-${agentName}-${date}-${random}`;
92
- }
93
- function createExperiment(agent, task, output, startedAt, completedAt, success, metrics, opts) {
94
- return {
95
- id: generateExperimentId(agent.name),
96
- agentName: agent.name,
97
- promptVersion: opts.promptVersion ?? 'v1',
98
- task,
99
- taskType: opts.taskType ?? DEFAULT_TASK_TYPE,
100
- startedAt,
101
- completedAt,
102
- success,
103
- metrics,
104
- output,
105
- };
106
- }
107
- // ─── Report Writer ───────────────────────────────────
108
- function saveReport(experiment, dataDir) {
109
- const reportsDir = resolve(join(dataDir, 'reports'));
110
- mkdirSync(reportsDir, { recursive: true });
111
- const filename = `${experiment.id}.md`;
112
- const filepath = join(reportsDir, filename);
113
- const durationSec = (experiment.metrics.durationMs / 1000).toFixed(1);
114
- const report = `# ${experiment.id}
115
-
116
- ## Meta
117
- - **Agent:** ${experiment.agentName}
118
- - **Task Type:** ${experiment.taskType}
119
- - **Prompt Version:** ${experiment.promptVersion}
120
- - **Started:** ${experiment.startedAt}
121
- - **Completed:** ${experiment.completedAt}
122
- - **Duration:** ${durationSec}s
123
- - **Success:** ${experiment.success ? 'Yes' : 'No'}
124
-
125
- ## Metrics
126
- | Metric | Value |
127
- |--------|-------|
128
- | Quality Score | ${experiment.metrics.qualityScore ?? 'pending'} |
129
- | Source Count | ${experiment.metrics.sourceCount} |
130
- | Output Length | ${experiment.metrics.outputLength} chars |
131
- | Error Count | ${experiment.metrics.errorCount} |
132
- | Duration | ${durationSec}s |
133
-
134
- ## Task
135
- ${experiment.task}
136
-
137
- ## Output
138
- ${experiment.output ?? 'No output captured'}
139
- `;
140
- writeFileSync(filepath, report, 'utf-8');
141
- return filepath;
142
- }
143
- // ─── Provider Resolution ─────────────────────────────
144
- /**
145
- * Resolve which LLM provider to use for a given agent and options.
146
- *
147
- * Priority: opts.provider > config.provider > 'claude-cli' (default)
148
- *
149
- * Auto-fallback: If agent uses MCP tools and provider doesn't support MCP,
150
- * automatically falls back to Claude CLI with a warning.
151
- */
152
- function resolveProvider(agent, opts) {
153
- // Explicit provider passed in options
154
- if (opts.provider) {
155
- return opts.provider;
156
- }
157
- // Determine provider type from config
158
- const providerType = opts.config?.provider ?? 'claude-cli';
159
- const provider = createProvider({ type: providerType });
160
- // Auto-fallback: if agent needs MCP but provider doesn't support it
161
- const needsMcp = agent.mcp && agent.mcp.length > 0;
162
- if (needsMcp && !provider.supportsMcp) {
163
- console.warn(`⚠ Agent "${agent.name}" uses MCP tools but provider "${provider.name}" doesn't support MCP. Falling back to claude-cli.`);
164
- return createProvider({ type: 'claude-cli' });
165
- }
166
- return provider;
167
- }
168
- // ─── Runner ──────────────────────────────────────────
169
- /**
170
- * Run an agent using the configured LLM provider.
171
- *
172
- * For Claude CLI: spawns a child process with MCP support.
173
- * For API providers: makes a direct HTTP call (10-100x faster).
174
- *
175
- * @example
176
- * ```ts
177
- * // Default: uses Claude CLI
178
- * const result = await runAgent(writer, 'Write a landing page', {
179
- * taskType: 'market',
180
- * config: darwinConfig,
181
- * });
182
- *
183
- * // Explicit provider: Anthropic API (faster, no MCP)
184
- * import { createProvider } from '../providers/index.js';
185
- * const result = await runAgent(critic, 'Evaluate this output', {
186
- * provider: createProvider({ type: 'anthropic-api' }),
187
- * });
188
- * ```
189
- */
190
- export async function runAgent(agent, task, opts = {}) {
191
- // System agents don't use LLM providers
192
- if (agent.type === 'system') {
193
- throw new Error(`Agent "${agent.name}" is a system agent. ` +
194
- `System agents must be run through their handler, not the CLI runner.`);
195
- }
196
- // Hard-stop before paying for the next provider call if the process has
197
- // blown through its budget (run count or wall-clock).
198
- enforceBudget();
199
- const startedAt = new Date().toISOString();
200
- const dataDir = opts.config?.dataDir ?? DEFAULT_DATA_DIR;
201
- const model = opts.model ?? agent.model ?? DEFAULT_MODEL;
202
- const timeout = opts.timeout ?? DEFAULT_TIMEOUT;
203
- runsThisProcess += 1;
204
- // Resolve provider
205
- const provider = resolveProvider(agent, opts);
206
- // Configure Claude CLI-specific options
207
- if (provider instanceof ClaudeCliProvider) {
208
- provider.setRunOptions({
209
- agent,
210
- darwinConfig: opts.config,
211
- maxTurns: opts.maxTurns ?? agent.maxTurns ?? DEFAULT_MAX_TURNS,
212
- cwd: opts.cwd,
213
- autonomous: opts.autonomous,
214
- });
215
- }
216
- // Execute via provider
217
- const result = await provider.run({
218
- model,
219
- systemPrompt: agent.systemPrompt,
220
- userMessage: task,
221
- timeout,
222
- });
223
- const output = result.output;
224
- const completedAt = new Date().toISOString();
225
- // Parse metrics from output
226
- const metrics = parseMetrics(output, result.durationMs);
227
- // Detect truncated runs (max turns reached = not a real success)
228
- const isMaxTurns = output.startsWith('Error: Reached max turns');
229
- const success = !isMaxTurns && output.length > 50;
230
- // Create experiment record
231
- const experiment = createExperiment(agent, task, output, startedAt, completedAt, success, metrics, opts);
232
- // Save report to disk
233
- const reportPath = saveReport(experiment, dataDir);
234
- return {
235
- experiment,
236
- output,
237
- reportPath,
238
- };
239
- }
240
- // ─── Error Type ──────────────────────────────────────
241
- /**
242
- * Error thrown when an LLM provider fails.
243
- * Preserves the exit code (CLI) and any partial output for debugging.
244
- */
245
- export class RunnerError extends Error {
246
- exitCode;
247
- partialOutput;
248
- constructor(message, exitCode, partialOutput) {
249
- super(message);
250
- this.exitCode = exitCode;
251
- this.partialOutput = partialOutput;
252
- this.name = 'RunnerError';
253
- }
254
- }
255
- //# sourceMappingURL=runner.js.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"runner.js","sourceRoot":"","sources":["../../src/core/runner.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAAE,SAAS,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AACnD,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAU1C,OAAO,EAAE,iBAAiB,EAAE,MAAM,4BAA4B,CAAC;AAC/D,OAAO,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAC;AA0BvD,wDAAwD;AAExD,MAAM,eAAe,GAAG,OAAO,CAAC,CAAC,aAAa;AAC9C,MAAM,aAAa,GAAG,0BAA0B,CAAC;AACjD,MAAM,iBAAiB,GAAG,EAAE,CAAC;AAC7B,MAAM,gBAAgB,GAAG,SAAS,CAAC;AACnC,MAAM,iBAAiB,GAAG,SAAS,CAAC;AAEpC,wDAAwD;AACxD,EAAE;AACF,yEAAyE;AACzE,2EAA2E;AAC3E,yEAAyE;AACzE,0EAA0E;AAC1E,yEAAyE;AACzE,uEAAuE;AACvE,yEAAyE;AAEzE,MAAM,4BAA4B,GAAG,GAAG,CAAC;AACzC,MAAM,uBAAuB,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,CAAC,MAAM;AAEtD,IAAI,iBAAiB,GAAG,MAAM,CAC5B,OAAO,CAAC,GAAG,CAAC,2BAA2B,IAAI,4BAA4B,CACxE,CAAC;AACF,IAAI,YAAY,GAAG,MAAM,CACvB,OAAO,CAAC,GAAG,CAAC,sBAAsB,IAAI,uBAAuB,CAC9D,CAAC;AACF,IAAI,eAAe,GAAG,CAAC,CAAC;AACxB,IAAI,cAAc,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;AAEhC,4EAA4E;AAC5E,MAAM,UAAU,oBAAoB,CAAC,CAAS;IAC5C,iBAAiB,GAAG,CAAC,CAAC;AACxB,CAAC;AACD,6EAA6E;AAC7E,MAAM,UAAU,eAAe,CAAC,EAAU;IACxC,YAAY,GAAG,EAAE,CAAC;AACpB,CAAC;AACD,6EAA6E;AAC7E,MAAM,UAAU,gBAAgB;IAC9B,eAAe,GAAG,CAAC,CAAC;IACpB,cAAc,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;AAC9B,CAAC;AAED,mEAAmE;AACnE,MAAM,OAAO,iBAAkB,SAAQ,KAAK;IACG;IAA7C,YAAY,OAAe,EAAkB,MAAuB;QAClE,KAAK,CAAC,OAAO,CAAC,CAAC;QAD4B,WAAM,GAAN,MAAM,CAAiB;QAElE,IAAI,CAAC,IAAI,GAAG,mBAAmB,CAAC;IAClC,CAAC;CACF;AAED,SAAS,aAAa;IACpB,IAAI,iBAAiB,GAAG,CAAC,IAAI,eAAe,IAAI,iBAAiB,EAAE,CAAC;QAClE,MAAM,IAAI,iBAAiB,CACzB,2BAA2B,MAAM,CAAC,eAAe,CAAC,IAAI,MAAM,CAAC,iBAAiB,CAAC,yBAAyB;YACxG,kEAAkE,EAClE,MAAM,CACP,CAAC;IACJ,CAAC;IACD,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,cAAc,CAAC;IAC5C,IAAI,YAAY,GAAG,CAAC,IAAI,OAAO,IAAI,YAAY,EAAE,CAAC;QAChD,MAAM,IAAI,iBAAiB,CACzB,kCAAkC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,GAAG,IAAI,CAAC,CAAC,OAAO,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,YAAY,GAAG,IAAI,CAAC,CAAC,KAAK;YACvH,kEAAkE,EAClE,MAAM,CACP,CAAC;IACJ,CAAC;AACH,CAAC;AAED,kDAAkD;AAClD,MAAM,WAAW,GAAG,2BAA2B,CAAC;AAEhD,wDAAwD;AAExD;;;;GAIG;AACH,SAAS,YAAY,CAAC,MAAc,EAAE,UAAkB;IACtD,MAAM,IAAI,GAAG,MAAM,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC;IACvC,MAAM,UAAU,GAAG,IAAI,CAAC,CAAC,CAAC,IAAI,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,GAAG,EAAU,CAAC;IAE5D,OAAO;QACL,YAAY,EAAE,IAAI,EAAQ,yBAAyB;QACnD,WAAW,EAAE,UAAU,CAAC,IAAI;QAC5B,YAAY,EAAE,MAAM,CAAC,MAAM;QAC3B,UAAU,EAAE,CAAC;QACb,UAAU;KACX,CAAC;AACJ,CAAC;AAED,wDAAwD;AAExD,SAAS,oBAAoB,CAAC,SAAiB;IAC7C,MAAM,IAAI,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IACnD,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IACvD,OAAO,OAAO,SAAS,IAAI,IAAI,IAAI,MAAM,EAAE,CAAC;AAC9C,CAAC;AAED,SAAS,gBAAgB,CACvB,KAAsB,EACtB,IAAY,EACZ,MAAc,EACd,SAAiB,EACjB,WAAmB,EACnB,OAAgB,EAChB,OAAsB,EACtB,IAAgB;IAEhB,OAAO;QACL,EAAE,EAAE,oBAAoB,CAAC,KAAK,CAAC,IAAI,CAAC;QACpC,SAAS,EAAE,KAAK,CAAC,IAAI;QACrB,aAAa,EAAE,IAAI,CAAC,aAAa,IAAI,IAAI;QACzC,IAAI;QACJ,QAAQ,EAAE,IAAI,CAAC,QAAQ,IAAI,iBAAiB;QAC5C,SAAS;QACT,WAAW;QACX,OAAO;QACP,OAAO;QACP,MAAM;KACP,CAAC;AACJ,CAAC;AAED,wDAAwD;AAExD,SAAS,UAAU,CACjB,UAA4B,EAC5B,OAAe;IAEf,MAAM,UAAU,GAAG,OAAO,CAAC,IAAI,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC;IACrD,SAAS,CAAC,UAAU,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAE3C,MAAM,QAAQ,GAAG,GAAG,UAAU,CAAC,EAAE,KAAK,CAAC;IACvC,MAAM,QAAQ,GAAG,IAAI,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAC;IAE5C,MAAM,WAAW,GAAG,CAAC,UAAU,CAAC,OAAO,CAAC,UAAU,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;IAEtE,MAAM,MAAM,GAAG,KAAK,UAAU,CAAC,EAAE;;;eAGpB,UAAU,CAAC,SAAS;mBAChB,UAAU,CAAC,QAAQ;wBACd,UAAU,CAAC,aAAa;iBAC/B,UAAU,CAAC,SAAS;mBAClB,UAAU,CAAC,WAAW;kBACvB,WAAW;iBACZ,UAAU,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI;;;;;oBAK9B,UAAU,CAAC,OAAO,CAAC,YAAY,IAAI,SAAS;mBAC7C,UAAU,CAAC,OAAO,CAAC,WAAW;oBAC7B,UAAU,CAAC,OAAO,CAAC,YAAY;kBACjC,UAAU,CAAC,OAAO,CAAC,UAAU;eAChC,WAAW;;;EAGxB,UAAU,CAAC,IAAI;;;EAGf,UAAU,CAAC,MAAM,IAAI,oBAAoB;CAC1C,CAAC;IAEA,aAAa,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC;IACzC,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,wDAAwD;AAExD;;;;;;;GAOG;AACH,SAAS,eAAe,CACtB,KAAsB,EACtB,IAAgB;IAEhB,sCAAsC;IACtC,IAAI,IAAI,CAAC,QAAQ,EAAE,CAAC;QAClB,OAAO,IAAI,CAAC,QAAQ,CAAC;IACvB,CAAC;IAED,sCAAsC;IACtC,MAAM,YAAY,GAAG,IAAI,CAAC,MAAM,EAAE,QAAQ,IAAI,YAAY,CAAC;IAE3D,MAAM,QAAQ,GAAG,cAAc,CAAC,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,CAAC;IAExD,oEAAoE;IACpE,MAAM,QAAQ,GAAG,KAAK,CAAC,GAAG,IAAI,KAAK,CAAC,GAAG,CAAC,MAAM,GAAG,CAAC,CAAC;IACnD,IAAI,QAAQ,IAAI,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC;QACtC,OAAO,CAAC,IAAI,CACV,YAAY,KAAK,CAAC,IAAI,kCAAkC,QAAQ,CAAC,IAAI,oDAAoD,CAC1H,CAAC;QACF,OAAO,cAAc,CAAC,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,CAAC;IAChD,CAAC;IAED,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,wDAAwD;AAExD;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,MAAM,CAAC,KAAK,UAAU,QAAQ,CAC5B,KAAsB,EACtB,IAAY,EACZ,OAAmB,EAAE;IAErB,wCAAwC;IACxC,IAAI,KAAK,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;QAC5B,MAAM,IAAI,KAAK,CACb,UAAU,KAAK,CAAC,IAAI,uBAAuB;YAC3C,sEAAsE,CACvE,CAAC;IACJ,CAAC;IAED,wEAAwE;IACxE,sDAAsD;IACtD,aAAa,EAAE,CAAC;IAEhB,MAAM,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC3C,MAAM,OAAO,GAAG,IAAI,CAAC,MAAM,EAAE,OAAO,IAAI,gBAAgB,CAAC;IACzD,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,KAAK,CAAC,KAAK,IAAI,aAAa,CAAC;IACzD,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,IAAI,eAAe,CAAC;IAEhD,eAAe,IAAI,CAAC,CAAC;IAErB,mBAAmB;IACnB,MAAM,QAAQ,GAAG,eAAe,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC;IAE9C,wCAAwC;IACxC,IAAI,QAAQ,YAAY,iBAAiB,EAAE,CAAC;QAC1C,QAAQ,CAAC,aAAa,CAAC;YACrB,KAAK;YACL,YAAY,EAAE,IAAI,CAAC,MAAM;YACzB,QAAQ,EAAE,IAAI,CAAC,QAAQ,IAAI,KAAK,CAAC,QAAQ,IAAI,iBAAiB;YAC9D,GAAG,EAAE,IAAI,CAAC,GAAG;YACb,UAAU,EAAE,IAAI,CAAC,UAAU;SAC5B,CAAC,CAAC;IACL,CAAC;IAED,uBAAuB;IACvB,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,GAAG,CAAC;QAChC,KAAK;QACL,YAAY,EAAE,KAAK,CAAC,YAAY;QAChC,WAAW,EAAE,IAAI;QACjB,OAAO;KACR,CAAC,CAAC;IAEH,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC;IAC7B,MAAM,WAAW,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAE7C,4BAA4B;IAC5B,MAAM,OAAO,GAAG,YAAY,CAAC,MAAM,EAAE,MAAM,CAAC,UAAU,CAAC,CAAC;IAExD,iEAAiE;IACjE,MAAM,UAAU,GAAG,MAAM,CAAC,UAAU,CAAC,0BAA0B,CAAC,CAAC;IACjE,MAAM,OAAO,GAAG,CAAC,UAAU,IAAI,MAAM,CAAC,MAAM,GAAG,EAAE,CAAC;IAElD,2BAA2B;IAC3B,MAAM,UAAU,GAAG,gBAAgB,CACjC,KAAK,EACL,IAAI,EACJ,MAAM,EACN,SAAS,EACT,WAAW,EACX,OAAO,EACP,OAAO,EACP,IAAI,CACL,CAAC;IAEF,sBAAsB;IACtB,MAAM,UAAU,GAAG,UAAU,CAAC,UAAU,EAAE,OAAO,CAAC,CAAC;IAEnD,OAAO;QACL,UAAU;QACV,MAAM;QACN,UAAU;KACX,CAAC;AACJ,CAAC;AAED,wDAAwD;AAExD;;;GAGG;AACH,MAAM,OAAO,WAAY,SAAQ,KAAK;IAGlB;IACA;IAHlB,YACE,OAAe,EACC,QAAuB,EACvB,aAAqB;QAErC,KAAK,CAAC,OAAO,CAAC,CAAC;QAHC,aAAQ,GAAR,QAAQ,CAAe;QACvB,kBAAa,GAAb,aAAa,CAAQ;QAGrC,IAAI,CAAC,IAAI,GAAG,aAAa,CAAC;IAC5B,CAAC;CACF"}
@@ -1,100 +0,0 @@
1
- /**
2
- * Darwin — Evolution Loop
3
- *
4
- * The core evolution cycle. Called after every agent run, it:
5
- * 1. Records the experiment
6
- * 2. Checks for failure rollback
7
- * 3. Manages A/B tests between prompt versions
8
- * 4. Triggers prompt optimization when enough data exists
9
- *
10
- * This is the brain of Darwin's self-evolution.
11
- */
12
- import type { AgentDefinition, DarwinExperiment, DarwinPattern, MemoryProvider } from '../types.js';
13
- import type { ExperimentTracker } from './tracker.js';
14
- import type { PromptOptimizer } from './optimizer.js';
15
- import type { SafetyGate } from './safety.js';
16
- import type { PatternDetector } from './patterns.js';
17
- import type { NotificationConfig } from './notifications.js';
18
- export interface EvolutionResult {
19
- patternsFound: DarwinPattern[];
20
- promptEvolved: boolean;
21
- abTestStarted: boolean;
22
- abTestCompleted: boolean;
23
- rolledBack: boolean;
24
- newVersion?: string;
25
- message: string;
26
- }
27
- interface DarwinLoopDeps {
28
- memory: MemoryProvider;
29
- tracker: ExperimentTracker;
30
- optimizer: PromptOptimizer;
31
- safety: SafetyGate;
32
- patterns: PatternDetector;
33
- /** Agent definition — used to pass tool context to optimizer */
34
- agent?: AgentDefinition;
35
- /** Notification config (Telegram alerts) — auto-loaded from env if not set */
36
- notifications?: NotificationConfig;
37
- }
38
- export declare class DarwinLoop {
39
- private memory;
40
- private tracker;
41
- private optimizer;
42
- private safety;
43
- private patterns;
44
- private agent?;
45
- private notifications;
46
- constructor(deps: DarwinLoopDeps);
47
- /**
48
- * Called AFTER every agent run. Drives the evolution cycle.
49
- *
50
- * Flow:
51
- * 0. Detect incomplete runs (skip them)
52
- * 1. Record experiment
53
- * 2. Rollback check (consecutive failures)
54
- * 3. A/B test management (if active)
55
- * 4. Validate data quality before evolving
56
- * 5. Evolution trigger (if enough data and no active test)
57
- */
58
- afterRun(experiment: DarwinExperiment): Promise<EvolutionResult>;
59
- private handleABTest;
60
- /**
61
- * Roll back to the last known good prompt version.
62
- * Returns true if a rollback was performed.
63
- */
64
- private rollback;
65
- /**
66
- * Activate a specific prompt version and deactivate all others.
67
- */
68
- private activateVersion;
69
- /**
70
- * Extract recent critic feedback reports from experiments.
71
- *
72
- * Returns up to `limit` feedback report texts from the most recent experiments
73
- * that have critic feedback. Experiments are already ordered by started_at DESC
74
- * from loadExperiments(), so we just filter for ones with feedback.
75
- */
76
- private getRecentFeedback;
77
- /**
78
- * Increment version string: "v1" -> "v2", "v12" -> "v13".
79
- */
80
- private nextVersion;
81
- /**
82
- * Check if a run is incomplete (agent ran out of turns or produced no real output).
83
- * Incomplete runs are NOT recorded as experiments to avoid poisoning the data.
84
- *
85
- * Checks output length regardless of success flag — a 300-char "successful" run
86
- * is still garbage data that shouldn't influence evolution.
87
- */
88
- private isIncompleteRun;
89
- /**
90
- * Validate that experiment data is clean enough for evolution.
91
- * Prevents garbage-in-garbage-out (e.g., broken search backend producing 0 sources).
92
- */
93
- private validateDataQuality;
94
- /**
95
- * Build a human-readable change reason from detected patterns.
96
- */
97
- private buildChangeReason;
98
- }
99
- export {};
100
- //# sourceMappingURL=loop.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"loop.d.ts","sourceRoot":"","sources":["../../src/evolution/loop.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,KAAK,EAEV,eAAe,EACf,gBAAgB,EAChB,aAAa,EACb,cAAc,EAEf,MAAM,aAAa,CAAC;AACrB,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,cAAc,CAAC;AACtD,OAAO,KAAK,EAAE,eAAe,EAAoB,MAAM,gBAAgB,CAAC;AACxE,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAC9C,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AACrD,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,oBAAoB,CAAC;AAK7D,MAAM,WAAW,eAAe;IAC9B,aAAa,EAAE,aAAa,EAAE,CAAC;IAC/B,aAAa,EAAE,OAAO,CAAC;IACvB,aAAa,EAAE,OAAO,CAAC;IACvB,eAAe,EAAE,OAAO,CAAC;IACzB,UAAU,EAAE,OAAO,CAAC;IACpB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC;CACjB;AAID,UAAU,cAAc;IACtB,MAAM,EAAE,cAAc,CAAC;IACvB,OAAO,EAAE,iBAAiB,CAAC;IAC3B,SAAS,EAAE,eAAe,CAAC;IAC3B,MAAM,EAAE,UAAU,CAAC;IACnB,QAAQ,EAAE,eAAe,CAAC;IAC1B,gEAAgE;IAChE,KAAK,CAAC,EAAE,eAAe,CAAC;IACxB,8EAA8E;IAC9E,aAAa,CAAC,EAAE,kBAAkB,CAAC;CACpC;AAYD,qBAAa,UAAU;IACrB,OAAO,CAAC,MAAM,CAAiB;IAC/B,OAAO,CAAC,OAAO,CAAoB;IACnC,OAAO,CAAC,SAAS,CAAkB;IACnC,OAAO,CAAC,MAAM,CAAa;IAC3B,OAAO,CAAC,QAAQ,CAAkB;IAClC,OAAO,CAAC,KAAK,CAAC,CAAkB;IAChC,OAAO,CAAC,aAAa,CAAqB;gBAE9B,IAAI,EAAE,cAAc;IAUhC;;;;;;;;;;OAUG;IACG,QAAQ,CAAC,UAAU,EAAE,gBAAgB,GAAG,OAAO,CAAC,eAAe,CAAC;YA4NxD,YAAY;IAiG1B;;;OAGG;YACW,QAAQ;IA8BtB;;OAEG;YACW,eAAe;IAe7B;;;;;;OAMG;YACW,iBAAiB;IAe/B;;OAEG;IACH,OAAO,CAAC,WAAW;IAWnB;;;;;;OAMG;IACH,OAAO,CAAC,eAAe;IAKvB;;;OAGG;YACW,mBAAmB;IAiDjC;;OAEG;IACH,OAAO,CAAC,iBAAiB;CAsB1B"}