darwin-agents 0.9.0 → 0.10.0
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.
- package/CHANGELOG.md +51 -0
- package/README.md +48 -0
- package/dist/src/cli/evolution-flags.d.ts +2 -0
- package/dist/src/cli/evolution-flags.d.ts.map +1 -1
- package/dist/src/cli/evolution-flags.js +36 -1
- package/dist/src/cli/evolution-flags.js.map +1 -1
- package/dist/src/cli/evolve.d.ts +2 -0
- package/dist/src/cli/evolve.d.ts.map +1 -1
- package/dist/src/cli/evolve.js +10 -0
- package/dist/src/cli/evolve.js.map +1 -1
- package/dist/src/cli/index.js +2 -0
- package/dist/src/cli/index.js.map +1 -1
- package/dist/src/evolution/demos.d.ts +128 -0
- package/dist/src/evolution/demos.d.ts.map +1 -0
- package/dist/src/evolution/demos.js +196 -0
- package/dist/src/evolution/demos.js.map +1 -0
- package/dist/src/evolution/enabled-state.d.ts +5 -5
- package/dist/src/evolution/enabled-state.d.ts.map +1 -1
- package/dist/src/evolution/enabled-state.js +10 -5
- package/dist/src/evolution/enabled-state.js.map +1 -1
- package/dist/src/evolution/loop.d.ts +36 -0
- package/dist/src/evolution/loop.d.ts.map +1 -1
- package/dist/src/evolution/loop.js +118 -19
- package/dist/src/evolution/loop.js.map +1 -1
- package/dist/src/evolution/selection.d.ts +73 -0
- package/dist/src/evolution/selection.d.ts.map +1 -0
- package/dist/src/evolution/selection.js +121 -0
- package/dist/src/evolution/selection.js.map +1 -0
- package/dist/src/index.d.ts +2 -0
- package/dist/src/index.d.ts.map +1 -1
- package/dist/src/index.js +10 -0
- package/dist/src/index.js.map +1 -1
- package/dist/src/types.d.ts +67 -0
- package/dist/src/types.d.ts.map +1 -1
- package/dist/src/types.js.map +1 -1
- package/package.json +4 -1
|
@@ -0,0 +1,196 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Demo injection (v0.10.0) — SIMBA-style "append a demo" challenger source.
|
|
3
|
+
*
|
|
4
|
+
* DSPy's SIMBA optimizer (Stochastic Introspective Mini-Batch Ascent) improves
|
|
5
|
+
* programs through two complementary strategies: appending self-reflective
|
|
6
|
+
* RULES (Darwin's reflector already covers that ground) and appending
|
|
7
|
+
* successful past examples as DEMONSTRATIONS. This module is Darwin's
|
|
8
|
+
* adaptation of the second strategy to the online evolution loop:
|
|
9
|
+
*
|
|
10
|
+
* 1. Harvest the agent's own highest-scoring past runs from memory
|
|
11
|
+
* (score ≥ threshold, successful, non-trivial output).
|
|
12
|
+
* 2. Prefer diversity: at most one demo per task type, best-scoring first.
|
|
13
|
+
* 3. Render them as a clearly-delimited "Demonstrations" section and
|
|
14
|
+
* append it to (or refresh it inside) the current prompt.
|
|
15
|
+
*
|
|
16
|
+
* The demo-augmented prompt is a CHALLENGER like any other — it goes through
|
|
17
|
+
* the same alignment guard, the same A/B test, and the same safety gate as a
|
|
18
|
+
* reflective or merged mutation. If demos don't actually help this agent, the
|
|
19
|
+
* incumbent wins and nothing changes.
|
|
20
|
+
*
|
|
21
|
+
* Design notes:
|
|
22
|
+
* - Zero LLM cost. Selection and rendering are pure string/array work on
|
|
23
|
+
* data Darwin already persists (task, output, score, taskType). This makes
|
|
24
|
+
* demos the cheapest challenger source in the loop.
|
|
25
|
+
* - Idempotent via markers. The section is wrapped in
|
|
26
|
+
* `<!-- darwin:demos:start/end -->` comments so a later demo cycle
|
|
27
|
+
* REPLACES the previous section instead of stacking a second one, and
|
|
28
|
+
* {@link stripDemoSection} can remove it cleanly.
|
|
29
|
+
* - Demo text comes from the agent's OWN prior outputs (already produced
|
|
30
|
+
* under this prompt lineage), not from an external input channel. Excerpts
|
|
31
|
+
* are length-capped and fenced, but callers embedding third-party data in
|
|
32
|
+
* tasks should be aware demos quote past tasks verbatim.
|
|
33
|
+
* - Determinism. Given the same experiment list and options, selection and
|
|
34
|
+
* rendering are fully deterministic (stable sort, no randomness) — the
|
|
35
|
+
* loop's no-op check ("would this produce the same prompt?") stays cheap.
|
|
36
|
+
*/
|
|
37
|
+
// ─── Defaults & markers ─────────────────────────────────────────────────────
|
|
38
|
+
export const DEFAULT_MAX_DEMOS = 2;
|
|
39
|
+
export const DEFAULT_DEMO_SCORE_THRESHOLD = 8;
|
|
40
|
+
export const DEFAULT_DEMO_MIN_OUTPUT_CHARS = 200;
|
|
41
|
+
export const DEFAULT_DEMO_TASK_CHARS = 240;
|
|
42
|
+
export const DEFAULT_DEMO_OUTPUT_CHARS = 900;
|
|
43
|
+
/** Start marker — everything between the markers is Darwin-managed. */
|
|
44
|
+
export const DEMO_SECTION_START = "<!-- darwin:demos:start -->";
|
|
45
|
+
/** End marker. */
|
|
46
|
+
export const DEMO_SECTION_END = "<!-- darwin:demos:end -->";
|
|
47
|
+
const DEMO_SECTION_INTRO = "## Demonstrations\n" +
|
|
48
|
+
"The following are excerpts from this agent's own highest-rated past " +
|
|
49
|
+
"responses. Match their quality, depth and structure — do NOT copy their " +
|
|
50
|
+
"content or treat their subject matter as part of the current task.";
|
|
51
|
+
// ─── Selection ──────────────────────────────────────────────────────────────
|
|
52
|
+
/** Clamp helper: finite number ≥ min, else fallback. */
|
|
53
|
+
function clampOption(value, min, fallback) {
|
|
54
|
+
return typeof value === "number" && Number.isFinite(value) && value >= min
|
|
55
|
+
? Math.floor(value)
|
|
56
|
+
: fallback;
|
|
57
|
+
}
|
|
58
|
+
/**
|
|
59
|
+
* Harvest demo candidates from an agent's experiment history.
|
|
60
|
+
*
|
|
61
|
+
* Filter: successful runs with a critic score ≥ `scoreThreshold` and an
|
|
62
|
+
* output of at least `minOutputChars`. Diversity: at most ONE demo per task
|
|
63
|
+
* type (the best-scoring run of that type, recency as tie-break) so two demos
|
|
64
|
+
* never showcase the same narrow strength. Order: score descending, then
|
|
65
|
+
* most-recent first — fully deterministic.
|
|
66
|
+
*/
|
|
67
|
+
export function selectDemoCandidates(experiments, options = {}) {
|
|
68
|
+
const maxDemos = clampOption(options.maxDemos, 1, DEFAULT_MAX_DEMOS);
|
|
69
|
+
const threshold = typeof options.scoreThreshold === "number" && Number.isFinite(options.scoreThreshold)
|
|
70
|
+
? options.scoreThreshold
|
|
71
|
+
: DEFAULT_DEMO_SCORE_THRESHOLD;
|
|
72
|
+
const minChars = clampOption(options.minOutputChars, 0, DEFAULT_DEMO_MIN_OUTPUT_CHARS);
|
|
73
|
+
const qualified = experiments.filter((exp) => {
|
|
74
|
+
if (!exp.success)
|
|
75
|
+
return false;
|
|
76
|
+
const score = exp.feedback?.score;
|
|
77
|
+
if (typeof score !== "number" || !Number.isFinite(score) || score < threshold)
|
|
78
|
+
return false;
|
|
79
|
+
const output = exp.output;
|
|
80
|
+
if (typeof output !== "string" || output.trim().length < minChars)
|
|
81
|
+
return false;
|
|
82
|
+
return typeof exp.task === "string" && exp.task.trim().length > 0;
|
|
83
|
+
});
|
|
84
|
+
// Best run per task type (score desc, then completedAt desc as tie-break).
|
|
85
|
+
const bestPerType = new Map();
|
|
86
|
+
for (const exp of qualified) {
|
|
87
|
+
const key = exp.taskType || "general";
|
|
88
|
+
const incumbent = bestPerType.get(key);
|
|
89
|
+
if (incumbent === undefined) {
|
|
90
|
+
bestPerType.set(key, exp);
|
|
91
|
+
continue;
|
|
92
|
+
}
|
|
93
|
+
const expScore = exp.feedback.score;
|
|
94
|
+
const incScore = incumbent.feedback.score;
|
|
95
|
+
if (expScore > incScore ||
|
|
96
|
+
(expScore === incScore && exp.completedAt > incumbent.completedAt)) {
|
|
97
|
+
bestPerType.set(key, exp);
|
|
98
|
+
}
|
|
99
|
+
}
|
|
100
|
+
const picked = [...bestPerType.values()].sort((a, b) => {
|
|
101
|
+
const scoreDelta = b.feedback.score - a.feedback.score;
|
|
102
|
+
if (scoreDelta !== 0)
|
|
103
|
+
return scoreDelta;
|
|
104
|
+
// Recency tie-break; ISO timestamps sort lexicographically.
|
|
105
|
+
if (a.completedAt < b.completedAt)
|
|
106
|
+
return 1;
|
|
107
|
+
if (a.completedAt > b.completedAt)
|
|
108
|
+
return -1;
|
|
109
|
+
return 0;
|
|
110
|
+
});
|
|
111
|
+
return picked.slice(0, maxDemos).map((exp) => ({
|
|
112
|
+
task: exp.task.trim(),
|
|
113
|
+
output: exp.output.trim(),
|
|
114
|
+
score: exp.feedback.score,
|
|
115
|
+
taskType: exp.taskType || "general",
|
|
116
|
+
}));
|
|
117
|
+
}
|
|
118
|
+
// ─── Rendering ──────────────────────────────────────────────────────────────
|
|
119
|
+
/**
|
|
120
|
+
* Strip the demo-section marker literals out of quoted demo text. A past
|
|
121
|
+
* output that itself contains `<!-- darwin:demos:end -->` (e.g. an agent
|
|
122
|
+
* writing ABOUT Darwin) would otherwise plant an early end-marker inside the
|
|
123
|
+
* section — the next refresh cycle's lazy marker-to-marker match would then
|
|
124
|
+
* cut at the wrong span and leave half the old section dangling in the prompt.
|
|
125
|
+
*/
|
|
126
|
+
function stripMarkerLiterals(text) {
|
|
127
|
+
return text.split(DEMO_SECTION_START).join("").split(DEMO_SECTION_END).join("");
|
|
128
|
+
}
|
|
129
|
+
/** Truncate at the last sentence/newline boundary within `maxLength`. */
|
|
130
|
+
function truncateAtSentenceBoundary(text, maxLength) {
|
|
131
|
+
if (text.length <= maxLength)
|
|
132
|
+
return text;
|
|
133
|
+
const truncated = text.slice(0, maxLength);
|
|
134
|
+
const lastPeriod = truncated.lastIndexOf(".");
|
|
135
|
+
const lastNewline = truncated.lastIndexOf("\n");
|
|
136
|
+
const cutPoint = Math.max(lastPeriod, lastNewline);
|
|
137
|
+
// Only honour the boundary when it keeps a meaningful chunk (> 40%).
|
|
138
|
+
const kept = cutPoint > maxLength * 0.4 ? truncated.slice(0, cutPoint + 1) : truncated;
|
|
139
|
+
return `${kept.trimEnd()} [...]`;
|
|
140
|
+
}
|
|
141
|
+
/**
|
|
142
|
+
* Render demo candidates as a marker-delimited prompt section.
|
|
143
|
+
*
|
|
144
|
+
* Returns an empty string for an empty demo list so callers can treat
|
|
145
|
+
* "nothing to inject" uniformly.
|
|
146
|
+
*/
|
|
147
|
+
export function buildDemoSection(demos, options = {}) {
|
|
148
|
+
if (demos.length === 0)
|
|
149
|
+
return "";
|
|
150
|
+
const maxTask = clampOption(options.maxTaskChars, 20, DEFAULT_DEMO_TASK_CHARS);
|
|
151
|
+
const maxOutput = clampOption(options.maxOutputChars, 80, DEFAULT_DEMO_OUTPUT_CHARS);
|
|
152
|
+
const blocks = demos.map((demo, i) => {
|
|
153
|
+
const task = truncateAtSentenceBoundary(stripMarkerLiterals(demo.task), maxTask);
|
|
154
|
+
const excerpt = truncateAtSentenceBoundary(stripMarkerLiterals(demo.output), maxOutput);
|
|
155
|
+
return (`### Demonstration ${i + 1} — task type "${demo.taskType}" (scored ${demo.score}/10)\n` +
|
|
156
|
+
`Task: ${task}\n` +
|
|
157
|
+
`Response excerpt:\n"""\n${excerpt}\n"""`);
|
|
158
|
+
});
|
|
159
|
+
return [DEMO_SECTION_START, DEMO_SECTION_INTRO, ...blocks, DEMO_SECTION_END].join("\n\n");
|
|
160
|
+
}
|
|
161
|
+
// ─── Application ────────────────────────────────────────────────────────────
|
|
162
|
+
/** Regex matching one marker-delimited demo section (including markers). */
|
|
163
|
+
const DEMO_SECTION_PATTERN = new RegExp(`${DEMO_SECTION_START.replace(/[.*+?^${}()|[\]\\]/g, "\\$&")}[\\s\\S]*?${DEMO_SECTION_END.replace(/[.*+?^${}()|[\]\\]/g, "\\$&")}`);
|
|
164
|
+
/**
|
|
165
|
+
* Apply a rendered demo section to a prompt.
|
|
166
|
+
*
|
|
167
|
+
* If the prompt already contains a marker-delimited demo section it is
|
|
168
|
+
* REPLACED in place (demos stay fresh, never stack); otherwise the section is
|
|
169
|
+
* appended after a blank line. An empty `section` returns the prompt with any
|
|
170
|
+
* existing demo section stripped — passing "no demos" acts as removal.
|
|
171
|
+
*
|
|
172
|
+
* The replacement uses a FUNCTION replacer: with a string replacer,
|
|
173
|
+
* `$&` / `$'` / `` $` `` / `$$` inside the section (an agent output quoting
|
|
174
|
+
* JS-regex or shell docs contains exactly these) would be interpreted as
|
|
175
|
+
* replacement patterns and garble the prompt on every refresh cycle.
|
|
176
|
+
*/
|
|
177
|
+
export function applyDemoSection(prompt, section) {
|
|
178
|
+
if (section === "")
|
|
179
|
+
return stripDemoSection(prompt);
|
|
180
|
+
if (DEMO_SECTION_PATTERN.test(prompt)) {
|
|
181
|
+
return prompt.replace(DEMO_SECTION_PATTERN, () => section);
|
|
182
|
+
}
|
|
183
|
+
const trimmed = prompt.replace(/\s+$/, "");
|
|
184
|
+
return `${trimmed}\n\n${section}`;
|
|
185
|
+
}
|
|
186
|
+
/**
|
|
187
|
+
* Remove the marker-delimited demo section (if any), collapsing the blank
|
|
188
|
+
* line that {@link applyDemoSection} inserted. Prompts without a section are
|
|
189
|
+
* returned unchanged.
|
|
190
|
+
*/
|
|
191
|
+
export function stripDemoSection(prompt) {
|
|
192
|
+
if (!DEMO_SECTION_PATTERN.test(prompt))
|
|
193
|
+
return prompt;
|
|
194
|
+
return prompt.replace(DEMO_SECTION_PATTERN, "").replace(/\n{3,}/g, "\n\n").replace(/\s+$/, "");
|
|
195
|
+
}
|
|
196
|
+
//# sourceMappingURL=demos.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"demos.js","sourceRoot":"","sources":["../../../src/evolution/demos.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAmCG;AAqDH,+EAA+E;AAE/E,MAAM,CAAC,MAAM,iBAAiB,GAAG,CAAC,CAAC;AACnC,MAAM,CAAC,MAAM,4BAA4B,GAAG,CAAC,CAAC;AAC9C,MAAM,CAAC,MAAM,6BAA6B,GAAG,GAAG,CAAC;AACjD,MAAM,CAAC,MAAM,uBAAuB,GAAG,GAAG,CAAC;AAC3C,MAAM,CAAC,MAAM,yBAAyB,GAAG,GAAG,CAAC;AAE7C,uEAAuE;AACvE,MAAM,CAAC,MAAM,kBAAkB,GAAG,6BAA6B,CAAC;AAChE,kBAAkB;AAClB,MAAM,CAAC,MAAM,gBAAgB,GAAG,2BAA2B,CAAC;AAE5D,MAAM,kBAAkB,GACtB,qBAAqB;IACrB,sEAAsE;IACtE,0EAA0E;IAC1E,oEAAoE,CAAC;AAEvE,+EAA+E;AAE/E,wDAAwD;AACxD,SAAS,WAAW,CAAC,KAAc,EAAE,GAAW,EAAE,QAAgB;IAChE,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,KAAK,IAAI,GAAG;QACxE,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC;QACnB,CAAC,CAAC,QAAQ,CAAC;AACf,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,oBAAoB,CAClC,WAA4C,EAC5C,UAAgC,EAAE;IAElC,MAAM,QAAQ,GAAG,WAAW,CAAC,OAAO,CAAC,QAAQ,EAAE,CAAC,EAAE,iBAAiB,CAAC,CAAC;IACrE,MAAM,SAAS,GACb,OAAO,OAAO,CAAC,cAAc,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,cAAc,CAAC;QACnF,CAAC,CAAC,OAAO,CAAC,cAAc;QACxB,CAAC,CAAC,4BAA4B,CAAC;IACnC,MAAM,QAAQ,GAAG,WAAW,CAAC,OAAO,CAAC,cAAc,EAAE,CAAC,EAAE,6BAA6B,CAAC,CAAC;IAEvF,MAAM,SAAS,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE;QAC3C,IAAI,CAAC,GAAG,CAAC,OAAO;YAAE,OAAO,KAAK,CAAC;QAC/B,MAAM,KAAK,GAAG,GAAG,CAAC,QAAQ,EAAE,KAAK,CAAC;QAClC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,SAAS;YAAE,OAAO,KAAK,CAAC;QAC5F,MAAM,MAAM,GAAG,GAAG,CAAC,MAAM,CAAC;QAC1B,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,MAAM,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,QAAQ;YAAE,OAAO,KAAK,CAAC;QAChF,OAAO,OAAO,GAAG,CAAC,IAAI,KAAK,QAAQ,IAAI,GAAG,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC;IACpE,CAAC,CAAC,CAAC;IAEH,2EAA2E;IAC3E,MAAM,WAAW,GAAG,IAAI,GAAG,EAA4B,CAAC;IACxD,KAAK,MAAM,GAAG,IAAI,SAAS,EAAE,CAAC;QAC5B,MAAM,GAAG,GAAG,GAAG,CAAC,QAAQ,IAAI,SAAS,CAAC;QACtC,MAAM,SAAS,GAAG,WAAW,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACvC,IAAI,SAAS,KAAK,SAAS,EAAE,CAAC;YAC5B,WAAW,CAAC,GAAG,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC;YAC1B,SAAS;QACX,CAAC;QACD,MAAM,QAAQ,GAAG,GAAG,CAAC,QAAS,CAAC,KAAK,CAAC;QACrC,MAAM,QAAQ,GAAG,SAAS,CAAC,QAAS,CAAC,KAAK,CAAC;QAC3C,IACE,QAAQ,GAAG,QAAQ;YACnB,CAAC,QAAQ,KAAK,QAAQ,IAAI,GAAG,CAAC,WAAW,GAAG,SAAS,CAAC,WAAW,CAAC,EAClE,CAAC;YACD,WAAW,CAAC,GAAG,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC;QAC5B,CAAC;IACH,CAAC;IAED,MAAM,MAAM,GAAG,CAAC,GAAG,WAAW,CAAC,MAAM,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;QACrD,MAAM,UAAU,GAAG,CAAC,CAAC,QAAS,CAAC,KAAK,GAAG,CAAC,CAAC,QAAS,CAAC,KAAK,CAAC;QACzD,IAAI,UAAU,KAAK,CAAC;YAAE,OAAO,UAAU,CAAC;QACxC,4DAA4D;QAC5D,IAAI,CAAC,CAAC,WAAW,GAAG,CAAC,CAAC,WAAW;YAAE,OAAO,CAAC,CAAC;QAC5C,IAAI,CAAC,CAAC,WAAW,GAAG,CAAC,CAAC,WAAW;YAAE,OAAO,CAAC,CAAC,CAAC;QAC7C,OAAO,CAAC,CAAC;IACX,CAAC,CAAC,CAAC;IAEH,OAAO,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC;QAC7C,IAAI,EAAE,GAAG,CAAC,IAAI,CAAC,IAAI,EAAE;QACrB,MAAM,EAAE,GAAG,CAAC,MAAO,CAAC,IAAI,EAAE;QAC1B,KAAK,EAAE,GAAG,CAAC,QAAS,CAAC,KAAK;QAC1B,QAAQ,EAAE,GAAG,CAAC,QAAQ,IAAI,SAAS;KACpC,CAAC,CAAC,CAAC;AACN,CAAC;AAED,+EAA+E;AAE/E;;;;;;GAMG;AACH,SAAS,mBAAmB,CAAC,IAAY;IACvC,OAAO,IAAI,CAAC,KAAK,CAAC,kBAAkB,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,gBAAgB,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;AAClF,CAAC;AAED,yEAAyE;AACzE,SAAS,0BAA0B,CAAC,IAAY,EAAE,SAAiB;IACjE,IAAI,IAAI,CAAC,MAAM,IAAI,SAAS;QAAE,OAAO,IAAI,CAAC;IAC1C,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC,CAAC;IAC3C,MAAM,UAAU,GAAG,SAAS,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC;IAC9C,MAAM,WAAW,GAAG,SAAS,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC;IAChD,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,CAAC,UAAU,EAAE,WAAW,CAAC,CAAC;IACnD,qEAAqE;IACrE,MAAM,IAAI,GAAG,QAAQ,GAAG,SAAS,GAAG,GAAG,CAAC,CAAC,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,EAAE,QAAQ,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACvF,OAAO,GAAG,IAAI,CAAC,OAAO,EAAE,QAAQ,CAAC;AACnC,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,gBAAgB,CAC9B,KAAmC,EACnC,UAA6B,EAAE;IAE/B,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAClC,MAAM,OAAO,GAAG,WAAW,CAAC,OAAO,CAAC,YAAY,EAAE,EAAE,EAAE,uBAAuB,CAAC,CAAC;IAC/E,MAAM,SAAS,GAAG,WAAW,CAAC,OAAO,CAAC,cAAc,EAAE,EAAE,EAAE,yBAAyB,CAAC,CAAC;IAErF,MAAM,MAAM,GAAG,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC,EAAE,EAAE;QACnC,MAAM,IAAI,GAAG,0BAA0B,CAAC,mBAAmB,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,OAAO,CAAC,CAAC;QACjF,MAAM,OAAO,GAAG,0BAA0B,CAAC,mBAAmB,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,SAAS,CAAC,CAAC;QACxF,OAAO,CACL,qBAAqB,CAAC,GAAG,CAAC,iBAAiB,IAAI,CAAC,QAAQ,aAAa,IAAI,CAAC,KAAK,QAAQ;YACvF,SAAS,IAAI,IAAI;YACjB,2BAA2B,OAAO,OAAO,CAC1C,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,OAAO,CAAC,kBAAkB,EAAE,kBAAkB,EAAE,GAAG,MAAM,EAAE,gBAAgB,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;AAC5F,CAAC;AAED,+EAA+E;AAE/E,4EAA4E;AAC5E,MAAM,oBAAoB,GAAG,IAAI,MAAM,CACrC,GAAG,kBAAkB,CAAC,OAAO,CAAC,qBAAqB,EAAE,MAAM,CAAC,aAAa,gBAAgB,CAAC,OAAO,CAAC,qBAAqB,EAAE,MAAM,CAAC,EAAE,CACnI,CAAC;AAEF;;;;;;;;;;;;GAYG;AACH,MAAM,UAAU,gBAAgB,CAAC,MAAc,EAAE,OAAe;IAC9D,IAAI,OAAO,KAAK,EAAE;QAAE,OAAO,gBAAgB,CAAC,MAAM,CAAC,CAAC;IACpD,IAAI,oBAAoB,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC;QACtC,OAAO,MAAM,CAAC,OAAO,CAAC,oBAAoB,EAAE,GAAG,EAAE,CAAC,OAAO,CAAC,CAAC;IAC7D,CAAC;IACD,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IAC3C,OAAO,GAAG,OAAO,OAAO,OAAO,EAAE,CAAC;AACpC,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,gBAAgB,CAAC,MAAc;IAC7C,IAAI,CAAC,oBAAoB,CAAC,IAAI,CAAC,MAAM,CAAC;QAAE,OAAO,MAAM,CAAC;IACtD,OAAO,MAAM,CAAC,OAAO,CAAC,oBAAoB,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;AACjG,CAAC"}
|
|
@@ -39,11 +39,11 @@ export declare function setEvolutionEnabled(memory: MemoryProvider, agentName: s
|
|
|
39
39
|
* ({@link DarwinState.evolutionConfigOverrides}) merged ON TOP, then any
|
|
40
40
|
* in-this-process CLI override merged last (e.g. `darwin run … --gepa`).
|
|
41
41
|
*
|
|
42
|
-
* Only the
|
|
43
|
-
* (`enabled`, `metrics`, `minRuns`, …) is taken verbatim from the
|
|
44
|
-
* config. Returns `undefined` when the agent has no `evolution` block
|
|
45
|
-
* persisted/CLI override would create one — overriding a flag on an
|
|
46
|
-
* never opted into evolution is meaningless.
|
|
42
|
+
* Only the advanced flags in {@link OVERRIDE_KEYS} are overridable; every
|
|
43
|
+
* other field (`enabled`, `metrics`, `minRuns`, …) is taken verbatim from the
|
|
44
|
+
* static config. Returns `undefined` when the agent has no `evolution` block
|
|
45
|
+
* AND no persisted/CLI override would create one — overriding a flag on an
|
|
46
|
+
* agent that never opted into evolution is meaningless.
|
|
47
47
|
*
|
|
48
48
|
* Read defensively: undefined override values are skipped so they never clobber
|
|
49
49
|
* a static default with `undefined`.
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"enabled-state.d.ts","sourceRoot":"","sources":["../../../src/evolution/enabled-state.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,KAAK,EACV,eAAe,EACf,WAAW,EACX,eAAe,EACf,uBAAuB,EACvB,cAAc,EACf,MAAM,aAAa,CAAC;AAErB;;;;;;;;;;;;GAYG;AACH,wBAAgB,uBAAuB,CACrC,KAAK,EAAE,eAAe,EACtB,KAAK,EAAE,IAAI,CAAC,WAAW,EAAE,kBAAkB,CAAC,GAC3C,OAAO,CAUT;AAED;;;;;;GAMG;AACH,wBAAsB,mBAAmB,CACvC,MAAM,EAAE,cAAc,EACtB,SAAS,EAAE,MAAM,EACjB,OAAO,EAAE,OAAO,GACf,OAAO,CAAC,IAAI,CAAC,CAQf;
|
|
1
|
+
{"version":3,"file":"enabled-state.d.ts","sourceRoot":"","sources":["../../../src/evolution/enabled-state.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,KAAK,EACV,eAAe,EACf,WAAW,EACX,eAAe,EACf,uBAAuB,EACvB,cAAc,EACf,MAAM,aAAa,CAAC;AAErB;;;;;;;;;;;;GAYG;AACH,wBAAgB,uBAAuB,CACrC,KAAK,EAAE,eAAe,EACtB,KAAK,EAAE,IAAI,CAAC,WAAW,EAAE,kBAAkB,CAAC,GAC3C,OAAO,CAUT;AAED;;;;;;GAMG;AACH,wBAAsB,mBAAmB,CACvC,MAAM,EAAE,cAAc,EACtB,SAAS,EAAE,MAAM,EACjB,OAAO,EAAE,OAAO,GACf,OAAO,CAAC,IAAI,CAAC,CAQf;AAeD;;;;;;;;;;;;;;GAcG;AACH,wBAAgB,sBAAsB,CACpC,KAAK,EAAE,eAAe,EACtB,KAAK,EAAE,IAAI,CAAC,WAAW,EAAE,0BAA0B,CAAC,EACpD,WAAW,CAAC,EAAE,uBAAuB,GACpC,eAAe,GAAG,SAAS,CAgC7B;AAED;;;;;GAKG;AACH,wBAAsB,2BAA2B,CAC/C,MAAM,EAAE,cAAc,EACtB,SAAS,EAAE,MAAM,EACjB,OAAO,EAAE,uBAAuB,GAC/B,OAAO,CAAC,IAAI,CAAC,CASf"}
|
|
@@ -58,6 +58,8 @@ const OVERRIDE_KEYS = [
|
|
|
58
58
|
'paretoGate',
|
|
59
59
|
'useCoverage',
|
|
60
60
|
'reflectionModel',
|
|
61
|
+
'useDemos',
|
|
62
|
+
'candidateSelection',
|
|
61
63
|
];
|
|
62
64
|
/**
|
|
63
65
|
* Resolve the effective {@link EvolutionConfig} for an agent: the static
|
|
@@ -65,11 +67,11 @@ const OVERRIDE_KEYS = [
|
|
|
65
67
|
* ({@link DarwinState.evolutionConfigOverrides}) merged ON TOP, then any
|
|
66
68
|
* in-this-process CLI override merged last (e.g. `darwin run … --gepa`).
|
|
67
69
|
*
|
|
68
|
-
* Only the
|
|
69
|
-
* (`enabled`, `metrics`, `minRuns`, …) is taken verbatim from the
|
|
70
|
-
* config. Returns `undefined` when the agent has no `evolution` block
|
|
71
|
-
* persisted/CLI override would create one — overriding a flag on an
|
|
72
|
-
* never opted into evolution is meaningless.
|
|
70
|
+
* Only the advanced flags in {@link OVERRIDE_KEYS} are overridable; every
|
|
71
|
+
* other field (`enabled`, `metrics`, `minRuns`, …) is taken verbatim from the
|
|
72
|
+
* static config. Returns `undefined` when the agent has no `evolution` block
|
|
73
|
+
* AND no persisted/CLI override would create one — overriding a flag on an
|
|
74
|
+
* agent that never opted into evolution is meaningless.
|
|
73
75
|
*
|
|
74
76
|
* Read defensively: undefined override values are skipped so they never clobber
|
|
75
77
|
* a static default with `undefined`.
|
|
@@ -95,6 +97,9 @@ export function resolveEvolutionConfig(agent, state, cliOverride) {
|
|
|
95
97
|
if (key === 'reflectionModel') {
|
|
96
98
|
resolved.reflectionModel = value;
|
|
97
99
|
}
|
|
100
|
+
else if (key === 'candidateSelection') {
|
|
101
|
+
resolved.candidateSelection = value;
|
|
102
|
+
}
|
|
98
103
|
else {
|
|
99
104
|
resolved[key] = value;
|
|
100
105
|
}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"enabled-state.js","sourceRoot":"","sources":["../../../src/evolution/enabled-state.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAUH;;;;;;;;;;;;GAYG;AACH,MAAM,UAAU,uBAAuB,CACrC,KAAsB,EACtB,KAA4C;IAE5C,MAAM,SAAS,GAAG,KAAK,CAAC,gBAAgB,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IACvD,IAAI,OAAO,SAAS,KAAK,SAAS,EAAE,CAAC;QACnC,0EAA0E;QAC1E,wEAAwE;QACxE,yEAAyE;QACzE,2EAA2E;QAC3E,OAAO,SAAS,IAAI,KAAK,CAAC,SAAS,KAAK,SAAS,CAAC;IACpD,CAAC;IACD,OAAO,KAAK,CAAC,SAAS,EAAE,OAAO,IAAI,KAAK,CAAC;AAC3C,CAAC;AAED;;;;;;GAMG;AACH,MAAM,CAAC,KAAK,UAAU,mBAAmB,CACvC,MAAsB,EACtB,SAAiB,EACjB,OAAgB;IAEhB,MAAM,MAAM,CAAC,WAAW,CAAC,CAAC,KAAK,EAAE,EAAE;QACjC,IAAI,CAAC,KAAK,CAAC,gBAAgB,EAAE,CAAC;YAC5B,KAAK,CAAC,gBAAgB,GAAG,EAAE,CAAC;QAC9B,CAAC;QACD,KAAK,CAAC,gBAAgB,CAAC,SAAS,CAAC,GAAG,OAAO,CAAC;QAC5C,OAAO,KAAK,CAAC;IACf,CAAC,CAAC,CAAC;AACL,CAAC;AAED,uEAAuE;AAEvE,yDAAyD;AACzD,MAAM,aAAa,GAAG;IACpB,SAAS;IACT,UAAU;IACV,YAAY;IACZ,aAAa;IACb,iBAAiB;
|
|
1
|
+
{"version":3,"file":"enabled-state.js","sourceRoot":"","sources":["../../../src/evolution/enabled-state.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAUH;;;;;;;;;;;;GAYG;AACH,MAAM,UAAU,uBAAuB,CACrC,KAAsB,EACtB,KAA4C;IAE5C,MAAM,SAAS,GAAG,KAAK,CAAC,gBAAgB,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IACvD,IAAI,OAAO,SAAS,KAAK,SAAS,EAAE,CAAC;QACnC,0EAA0E;QAC1E,wEAAwE;QACxE,yEAAyE;QACzE,2EAA2E;QAC3E,OAAO,SAAS,IAAI,KAAK,CAAC,SAAS,KAAK,SAAS,CAAC;IACpD,CAAC;IACD,OAAO,KAAK,CAAC,SAAS,EAAE,OAAO,IAAI,KAAK,CAAC;AAC3C,CAAC;AAED;;;;;;GAMG;AACH,MAAM,CAAC,KAAK,UAAU,mBAAmB,CACvC,MAAsB,EACtB,SAAiB,EACjB,OAAgB;IAEhB,MAAM,MAAM,CAAC,WAAW,CAAC,CAAC,KAAK,EAAE,EAAE;QACjC,IAAI,CAAC,KAAK,CAAC,gBAAgB,EAAE,CAAC;YAC5B,KAAK,CAAC,gBAAgB,GAAG,EAAE,CAAC;QAC9B,CAAC;QACD,KAAK,CAAC,gBAAgB,CAAC,SAAS,CAAC,GAAG,OAAO,CAAC;QAC5C,OAAO,KAAK,CAAC;IACf,CAAC,CAAC,CAAC;AACL,CAAC;AAED,uEAAuE;AAEvE,yDAAyD;AACzD,MAAM,aAAa,GAAG;IACpB,SAAS;IACT,UAAU;IACV,YAAY;IACZ,aAAa;IACb,iBAAiB;IACjB,UAAU;IACV,oBAAoB;CACZ,CAAC;AAEX;;;;;;;;;;;;;;GAcG;AACH,MAAM,UAAU,sBAAsB,CACpC,KAAsB,EACtB,KAAoD,EACpD,WAAqC;IAErC,MAAM,SAAS,GAAG,KAAK,CAAC,wBAAwB,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAC/D,MAAM,IAAI,GAAG,KAAK,CAAC,SAAS,CAAC;IAE7B,IAAI,CAAC,IAAI,IAAI,CAAC,SAAS,IAAI,CAAC,WAAW,EAAE,CAAC;QACxC,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,2EAA2E;IAC3E,2EAA2E;IAC3E,6EAA6E;IAC7E,MAAM,QAAQ,GAAoB,IAAI,CAAC,CAAC,CAAC,EAAE,GAAG,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;IAE1E,KAAK,MAAM,KAAK,IAAI,CAAC,SAAS,EAAE,WAAW,CAAC,EAAE,CAAC;QAC7C,IAAI,CAAC,KAAK;YAAE,SAAS;QACrB,KAAK,MAAM,GAAG,IAAI,aAAa,EAAE,CAAC;YAChC,MAAM,KAAK,GAAG,KAAK,CAAC,GAAG,CAAC,CAAC;YACzB,IAAI,KAAK,KAAK,SAAS,EAAE,CAAC;gBACxB,oEAAoE;gBACpE,kCAAkC;gBAClC,IAAI,GAAG,KAAK,iBAAiB,EAAE,CAAC;oBAC9B,QAAQ,CAAC,eAAe,GAAG,KAAe,CAAC;gBAC7C,CAAC;qBAAM,IAAI,GAAG,KAAK,oBAAoB,EAAE,CAAC;oBACxC,QAAQ,CAAC,kBAAkB,GAAG,KAA8C,CAAC;gBAC/E,CAAC;qBAAM,CAAC;oBACN,QAAQ,CAAC,GAAG,CAAC,GAAG,KAAgB,CAAC;gBACnC,CAAC;YACH,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED;;;;;GAKG;AACH,MAAM,CAAC,KAAK,UAAU,2BAA2B,CAC/C,MAAsB,EACtB,SAAiB,EACjB,OAAgC;IAEhC,MAAM,MAAM,CAAC,WAAW,CAAC,CAAC,KAAK,EAAE,EAAE;QACjC,IAAI,CAAC,KAAK,CAAC,wBAAwB,EAAE,CAAC;YACpC,KAAK,CAAC,wBAAwB,GAAG,EAAE,CAAC;QACtC,CAAC;QACD,MAAM,QAAQ,GAAG,KAAK,CAAC,wBAAwB,CAAC,SAAS,CAAC,IAAI,EAAE,CAAC;QACjE,KAAK,CAAC,wBAAwB,CAAC,SAAS,CAAC,GAAG,EAAE,GAAG,QAAQ,EAAE,GAAG,OAAO,EAAE,CAAC;QACxE,OAAO,KAAK,CAAC;IACf,CAAC,CAAC,CAAC;AACL,CAAC"}
|
|
@@ -58,6 +58,13 @@ interface DarwinLoopDeps {
|
|
|
58
58
|
* (only used when `embed` is set). Default 0.82.
|
|
59
59
|
*/
|
|
60
60
|
alignmentSimilarityThreshold?: number;
|
|
61
|
+
/**
|
|
62
|
+
* v0.10.0 — Random source in [0, 1) for the stochastic parent-selection
|
|
63
|
+
* strategies (`candidateSelection: 'pareto' | 'epsilon-greedy'`). Injected
|
|
64
|
+
* for deterministic tests; default `Math.random`. Never consulted on the
|
|
65
|
+
* default `'active'` path.
|
|
66
|
+
*/
|
|
67
|
+
rng?: () => number;
|
|
61
68
|
}
|
|
62
69
|
export declare class DarwinLoop {
|
|
63
70
|
private memory;
|
|
@@ -70,6 +77,7 @@ export declare class DarwinLoop {
|
|
|
70
77
|
private gepa?;
|
|
71
78
|
private embed?;
|
|
72
79
|
private alignmentSimilarityThreshold?;
|
|
80
|
+
private rng?;
|
|
73
81
|
constructor(deps: DarwinLoopDeps);
|
|
74
82
|
/**
|
|
75
83
|
* Called AFTER every agent run. Drives the evolution cycle.
|
|
@@ -173,6 +181,34 @@ export declare class DarwinLoop {
|
|
|
173
181
|
* v0.7.0 — Merge cadence (every K-th cycle). Default 3, clamped ≥ 1.
|
|
174
182
|
*/
|
|
175
183
|
private mergeEveryK;
|
|
184
|
+
/**
|
|
185
|
+
* v0.10.0 — Demo-injection cadence (every K-th cycle). Default 4, clamped
|
|
186
|
+
* ≥ 1. Deliberately offset from the merge default (3) so the two
|
|
187
|
+
* non-reflective challenger sources don't collide on the same cycles.
|
|
188
|
+
*/
|
|
189
|
+
private demoEveryK;
|
|
190
|
+
/**
|
|
191
|
+
* v0.10.0 — Build a SIMBA-style demo challenger: current prompt + a
|
|
192
|
+
* marker-delimited "Demonstrations" section harvested from this agent's
|
|
193
|
+
* highest-scoring past runs. Pure selection + rendering — no LLM call.
|
|
194
|
+
*
|
|
195
|
+
* Returns `null` (→ caller falls through to GEPA/legacy generation) when
|
|
196
|
+
* no run qualifies (score/threshold/length filters) or when the rendered
|
|
197
|
+
* demo set is byte-identical to what the prompt already carries — an A/B
|
|
198
|
+
* test of a version against itself would be pointless.
|
|
199
|
+
*/
|
|
200
|
+
private tryDemoVariant;
|
|
201
|
+
/**
|
|
202
|
+
* v0.10.0 — Scored version history for parent selection AND merge: every
|
|
203
|
+
* prompt version that has text and at least one run's worth of averaged
|
|
204
|
+
* metrics, as {@link ScoredVariant}s. One getAverageMetrics
|
|
205
|
+
* (→ loadExperiments) call per version. O(N) backend round-trips, but N is
|
|
206
|
+
* the agent's prompt-version count (≤ ~20 in practice) and this only runs
|
|
207
|
+
* on selection/merge cadences, so it is cheap — same pattern handleABTest
|
|
208
|
+
* already uses for the Pareto gate. (Extracted from tryMergeVariant,
|
|
209
|
+
* behaviour unchanged.)
|
|
210
|
+
*/
|
|
211
|
+
private buildScoredHistory;
|
|
176
212
|
/**
|
|
177
213
|
* v0.7.0 — GEPA system-aware MERGE (paper Appendix-D). Builds scored
|
|
178
214
|
* variants from this agent's prompt-version history (each version's prompt
|
|
@@ -1 +1 @@
|
|
|
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,EAGf,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,EAAiB,MAAM,aAAa,CAAC;AAC7D,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AACrD,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,oBAAoB,CAAC;AAE7D,OAAO,KAAK,EAAE,aAAa,EAAiB,MAAM,qBAAqB,CAAC;AAGxE,OAAO,EAAkE,KAAK,OAAO,EAAE,MAAM,gBAAgB,CAAC;
|
|
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,EAGf,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,EAAiB,MAAM,aAAa,CAAC;AAC7D,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AACrD,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,oBAAoB,CAAC;AAE7D,OAAO,KAAK,EAAE,aAAa,EAAiB,MAAM,qBAAqB,CAAC;AAGxE,OAAO,EAAkE,KAAK,OAAO,EAAE,MAAM,gBAAgB,CAAC;AAO9G,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;IACnC;;;;;;OAMG;IACH,IAAI,CAAC,EAAE,aAAa,CAAC;IACrB;;;;;;;OAOG;IACH,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB;;;OAGG;IACH,4BAA4B,CAAC,EAAE,MAAM,CAAC;IACtC;;;;;OAKG;IACH,GAAG,CAAC,EAAE,MAAM,MAAM,CAAC;CACpB;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;IAC1C,OAAO,CAAC,IAAI,CAAC,CAAgB;IAC7B,OAAO,CAAC,KAAK,CAAC,CAAU;IACxB,OAAO,CAAC,4BAA4B,CAAC,CAAS;IAC9C,OAAO,CAAC,GAAG,CAAC,CAAe;gBAEf,IAAI,EAAE,cAAc;IAchC;;;;;;;;;;OAUG;IACG,QAAQ,CAAC,UAAU,EAAE,gBAAgB,GAAG,OAAO,CAAC,eAAe,CAAC;IAuJtE;;;;;;;;;;;;;;;;;;OAkBG;IACG,WAAW,CAAC,SAAS,EAAE,MAAM,GAAG,OAAO,CAAC,eAAe,CAAC;IAqC9D;;;;;;OAMG;YACW,sBAAsB;YA8ItB,YAAY;IAwJ1B;;;OAGG;YACW,QAAQ;IA8BtB;;OAEG;YACW,eAAe;IAe7B;;;;OAIG;IACH,OAAO,CAAC,cAAc;IAMtB;;;OAGG;IACH,OAAO,CAAC,UAAU;IAKlB;;;;;;OAMG;YACW,iBAAiB;IAe/B;;;;;;;;;;;;;;;;;OAiBG;YACW,mBAAmB;IAkHjC;;;;;;;;OAQG;YACW,iBAAiB;IAa/B;;OAEG;IACH,OAAO,CAAC,WAAW;IAMnB;;;;OAIG;IACH,OAAO,CAAC,UAAU;IAMlB;;;;;;;;;OASG;YACW,cAAc;IAa5B;;;;;;;;;OASG;YACW,kBAAkB;IAYhC;;;;;;;;;;;;OAYG;YACW,eAAe;IAgC7B;;;;;;;;;;;;;;;;OAgBG;YACW,oBAAoB;IAyBlC;;;;;;OAMG;YACW,qBAAqB;IAsBnC;;OAEG;IACH,OAAO,CAAC,WAAW;IAWnB;;;;;;OAMG;IACH,OAAO,CAAC,eAAe;IAKvB;;;OAGG;YACW,mBAAmB;IAiDjC;;OAEG;IACH,OAAO,CAAC,iBAAiB;CAsB1B"}
|
|
@@ -13,6 +13,8 @@ import { notifyABTestComplete, notifyEvolutionStarted, notifyRollback } from './
|
|
|
13
13
|
import { epochShuffledMinibatch } from './optimizer-gepa.js';
|
|
14
14
|
import { checkAlignmentPreservation, checkAlignmentPreservationSemantic } from './alignment.js';
|
|
15
15
|
import { dominatesEpsilon, paretoSelect, DARWIN_DEFAULT_OBJECTIVES } from './pareto.js';
|
|
16
|
+
import { selectDemoCandidates, buildDemoSection, applyDemoSection } from './demos.js';
|
|
17
|
+
import { selectParentVariant } from './selection.js';
|
|
16
18
|
// ─── Validation Constants ─────────────────────────────
|
|
17
19
|
/** Default minimum output length (overridden by agent.evolution.minOutputLength) */
|
|
18
20
|
const DEFAULT_MIN_VALID_OUTPUT = 2000;
|
|
@@ -30,6 +32,7 @@ export class DarwinLoop {
|
|
|
30
32
|
gepa;
|
|
31
33
|
embed;
|
|
32
34
|
alignmentSimilarityThreshold;
|
|
35
|
+
rng;
|
|
33
36
|
constructor(deps) {
|
|
34
37
|
this.memory = deps.memory;
|
|
35
38
|
this.tracker = deps.tracker;
|
|
@@ -41,6 +44,7 @@ export class DarwinLoop {
|
|
|
41
44
|
this.gepa = deps.gepa;
|
|
42
45
|
this.embed = deps.embed;
|
|
43
46
|
this.alignmentSimilarityThreshold = deps.alignmentSimilarityThreshold;
|
|
47
|
+
this.rng = deps.rng;
|
|
44
48
|
}
|
|
45
49
|
/**
|
|
46
50
|
* Called AFTER every agent run. Drives the evolution cycle.
|
|
@@ -250,7 +254,33 @@ export class DarwinLoop {
|
|
|
250
254
|
// optimizer so the loop never stalls.
|
|
251
255
|
let newPromptText = null;
|
|
252
256
|
let generatedBy = 'legacy';
|
|
253
|
-
|
|
257
|
+
// ── v0.10.0: SIMBA-style demo injection (opt-in, zero LLM cost) ──
|
|
258
|
+
// On every demoEveryK-th cycle, build a challenger by appending (or
|
|
259
|
+
// refreshing) a "Demonstrations" section harvested from the agent's own
|
|
260
|
+
// highest-scoring past runs. Independent of useGepa — no reflector is
|
|
261
|
+
// involved. The demo prompt runs the same alignment guard and enters the
|
|
262
|
+
// same A/B test as any mutation; when demos don't help, the incumbent
|
|
263
|
+
// wins. When the demo set is unchanged since the last injection the path
|
|
264
|
+
// yields no challenger (no-op) and falls through to GEPA/legacy.
|
|
265
|
+
if (this.agent?.evolution?.useDemos === true) {
|
|
266
|
+
const epoch = this.versionInt(activePrompt.version);
|
|
267
|
+
if (epoch > 0 && epoch % this.demoEveryK() === 0) {
|
|
268
|
+
const demoPrompt = await this.tryDemoVariant(agent, activePrompt.promptText);
|
|
269
|
+
if (demoPrompt !== null) {
|
|
270
|
+
const guarded = await this.runAlignmentGuard(activePrompt.promptText, demoPrompt);
|
|
271
|
+
if (guarded !== null) {
|
|
272
|
+
newPromptText = guarded;
|
|
273
|
+
generatedBy = 'demos';
|
|
274
|
+
}
|
|
275
|
+
else {
|
|
276
|
+
// Mirrors the merge-path breadcrumb: a consistently guard-failing
|
|
277
|
+
// demo section should be visible, not silently skipped forever.
|
|
278
|
+
console.warn(`[darwin] demo variant for "${agent}" failed the alignment guard, falling through.`);
|
|
279
|
+
}
|
|
280
|
+
}
|
|
281
|
+
}
|
|
282
|
+
}
|
|
283
|
+
if (newPromptText === null && this.agent?.evolution?.useGepa === true && this.gepa) {
|
|
254
284
|
// Epoch = the integer of the active prompt version (v1→1, v12→12). It
|
|
255
285
|
// advances by one each evolution cycle, so the epoch-shuffled minibatch
|
|
256
286
|
// rotates which feedback subset the reflector sees per cycle, and the
|
|
@@ -273,10 +303,11 @@ export class DarwinLoop {
|
|
|
273
303
|
promptText: newPromptText,
|
|
274
304
|
createdAt: new Date().toISOString(),
|
|
275
305
|
parentVersion: activePrompt.version,
|
|
276
|
-
// Only tag the change reason with the generator when
|
|
277
|
-
//
|
|
306
|
+
// Only tag the change reason with the generator when a non-legacy
|
|
307
|
+
// challenger source was opted in (GEPA v0.6.0 / demos v0.10.0) — keeps
|
|
308
|
+
// the stored `changeReason` byte-for-byte identical for legacy-only
|
|
278
309
|
// agents (v0.6.0 review Finding 8).
|
|
279
|
-
changeReason: this.agent?.evolution?.useGepa
|
|
310
|
+
changeReason: this.agent?.evolution?.useGepa || this.agent?.evolution?.useDemos
|
|
280
311
|
? `[${generatedBy}] ${this.buildChangeReason(detectedPatterns)}`
|
|
281
312
|
: this.buildChangeReason(detectedPatterns),
|
|
282
313
|
active: false, // Not active yet — going into A/B test
|
|
@@ -305,7 +336,9 @@ export class DarwinLoop {
|
|
|
305
336
|
result.promptEvolved = true;
|
|
306
337
|
result.abTestStarted = true;
|
|
307
338
|
result.newVersion = newVersion;
|
|
308
|
-
const genLabel = this.agent?.evolution?.useGepa
|
|
339
|
+
const genLabel = this.agent?.evolution?.useGepa || this.agent?.evolution?.useDemos
|
|
340
|
+
? ` via ${generatedBy}`
|
|
341
|
+
: '';
|
|
309
342
|
result.message = `New prompt ${newVersion} generated${genLabel}. A/B test started: ${activePrompt.version} vs ${newVersion} (minRuns: ${dynamicMinRuns}).`;
|
|
310
343
|
// Notify via Telegram (non-blocking)
|
|
311
344
|
notifyEvolutionStarted(this.notifications, agent, activePrompt.version, newVersion, this.buildChangeReason(detectedPatterns)).catch(() => { });
|
|
@@ -554,10 +587,31 @@ export class DarwinLoop {
|
|
|
554
587
|
// currently-active prompt. Falls back to the active prompt when fewer than
|
|
555
588
|
// two versions carry per-task-type data (selectCoverageParent → null).
|
|
556
589
|
let parentPrompt = currentPrompt;
|
|
590
|
+
let parentChosen = false;
|
|
557
591
|
if (this.agent?.evolution?.useCoverage === true) {
|
|
558
592
|
const coverageParent = await this.selectCoverageParent(agentName);
|
|
559
593
|
if (coverageParent !== null) {
|
|
560
594
|
parentPrompt = coverageParent;
|
|
595
|
+
parentChosen = true;
|
|
596
|
+
}
|
|
597
|
+
}
|
|
598
|
+
// v0.10.0: GEPA candidate_selection_strategy parity (opt-in via
|
|
599
|
+
// evolution.candidateSelection). Pick the reflection parent from the
|
|
600
|
+
// agent's SCORED version history — 'best' (current_best), 'pareto'
|
|
601
|
+
// (uniform sample from the non-dominated front), or 'epsilon-greedy'
|
|
602
|
+
// (explore/exploit). Coverage selection (the more specific GEPA
|
|
603
|
+
// Algorithm 2 selector) takes precedence when it found a parent; this
|
|
604
|
+
// strategy is the fallback for that cycle. 'active'/unset keeps the
|
|
605
|
+
// historical reflect-from-active-prompt behaviour byte-for-byte.
|
|
606
|
+
const strategy = this.agent?.evolution?.candidateSelection;
|
|
607
|
+
if (!parentChosen && strategy !== undefined && strategy !== 'active') {
|
|
608
|
+
const scored = await this.buildScoredHistory(agentName);
|
|
609
|
+
const chosen = selectParentVariant(scored, strategy, {
|
|
610
|
+
epsilon: this.agent?.evolution?.explorationEpsilon,
|
|
611
|
+
rng: this.rng,
|
|
612
|
+
});
|
|
613
|
+
if (chosen !== null) {
|
|
614
|
+
parentPrompt = chosen.prompt;
|
|
561
615
|
}
|
|
562
616
|
}
|
|
563
617
|
// v0.7.0: pull the configurable feedback window (default 15, was 5).
|
|
@@ -625,6 +679,64 @@ export class DarwinLoop {
|
|
|
625
679
|
return Math.floor(k);
|
|
626
680
|
return 3;
|
|
627
681
|
}
|
|
682
|
+
/**
|
|
683
|
+
* v0.10.0 — Demo-injection cadence (every K-th cycle). Default 4, clamped
|
|
684
|
+
* ≥ 1. Deliberately offset from the merge default (3) so the two
|
|
685
|
+
* non-reflective challenger sources don't collide on the same cycles.
|
|
686
|
+
*/
|
|
687
|
+
demoEveryK() {
|
|
688
|
+
const k = this.agent?.evolution?.demoEveryK;
|
|
689
|
+
if (typeof k === 'number' && Number.isFinite(k) && k >= 1)
|
|
690
|
+
return Math.floor(k);
|
|
691
|
+
return 4;
|
|
692
|
+
}
|
|
693
|
+
/**
|
|
694
|
+
* v0.10.0 — Build a SIMBA-style demo challenger: current prompt + a
|
|
695
|
+
* marker-delimited "Demonstrations" section harvested from this agent's
|
|
696
|
+
* highest-scoring past runs. Pure selection + rendering — no LLM call.
|
|
697
|
+
*
|
|
698
|
+
* Returns `null` (→ caller falls through to GEPA/legacy generation) when
|
|
699
|
+
* no run qualifies (score/threshold/length filters) or when the rendered
|
|
700
|
+
* demo set is byte-identical to what the prompt already carries — an A/B
|
|
701
|
+
* test of a version against itself would be pointless.
|
|
702
|
+
*/
|
|
703
|
+
async tryDemoVariant(agentName, currentPrompt) {
|
|
704
|
+
const experiments = await this.memory.loadExperiments(agentName);
|
|
705
|
+
const demos = selectDemoCandidates(experiments, {
|
|
706
|
+
maxDemos: this.agent?.evolution?.maxDemos,
|
|
707
|
+
scoreThreshold: this.agent?.evolution?.demoScoreThreshold,
|
|
708
|
+
});
|
|
709
|
+
if (demos.length === 0)
|
|
710
|
+
return null;
|
|
711
|
+
const section = buildDemoSection(demos);
|
|
712
|
+
const withDemos = applyDemoSection(currentPrompt, section);
|
|
713
|
+
if (withDemos === currentPrompt)
|
|
714
|
+
return null;
|
|
715
|
+
return withDemos;
|
|
716
|
+
}
|
|
717
|
+
/**
|
|
718
|
+
* v0.10.0 — Scored version history for parent selection AND merge: every
|
|
719
|
+
* prompt version that has text and at least one run's worth of averaged
|
|
720
|
+
* metrics, as {@link ScoredVariant}s. One getAverageMetrics
|
|
721
|
+
* (→ loadExperiments) call per version. O(N) backend round-trips, but N is
|
|
722
|
+
* the agent's prompt-version count (≤ ~20 in practice) and this only runs
|
|
723
|
+
* on selection/merge cadences, so it is cheap — same pattern handleABTest
|
|
724
|
+
* already uses for the Pareto gate. (Extracted from tryMergeVariant,
|
|
725
|
+
* behaviour unchanged.)
|
|
726
|
+
*/
|
|
727
|
+
async buildScoredHistory(agentName) {
|
|
728
|
+
const versions = await this.memory.getAllPromptVersions(agentName);
|
|
729
|
+
const scored = [];
|
|
730
|
+
for (const v of versions) {
|
|
731
|
+
if (!v.promptText)
|
|
732
|
+
continue;
|
|
733
|
+
const metrics = await this.tracker.getAverageMetrics(agentName, v.version);
|
|
734
|
+
if (Object.keys(metrics).length === 0)
|
|
735
|
+
continue; // no runs for this version
|
|
736
|
+
scored.push({ id: v.version, prompt: v.promptText, metrics });
|
|
737
|
+
}
|
|
738
|
+
return scored;
|
|
739
|
+
}
|
|
628
740
|
/**
|
|
629
741
|
* v0.7.0 — GEPA system-aware MERGE (paper Appendix-D). Builds scored
|
|
630
742
|
* variants from this agent's prompt-version history (each version's prompt
|
|
@@ -641,20 +753,7 @@ export class DarwinLoop {
|
|
|
641
753
|
async tryMergeVariant(agentName) {
|
|
642
754
|
if (!this.gepa)
|
|
643
755
|
return null;
|
|
644
|
-
const
|
|
645
|
-
const scored = [];
|
|
646
|
-
// One getAverageMetrics (→ loadExperiments) call per version. O(N) backend
|
|
647
|
-
// round-trips, but N is the agent's prompt-version count (≤ ~20 in
|
|
648
|
-
// practice) and this only runs on the merge cadence, so it is cheap — same
|
|
649
|
-
// pattern handleABTest already uses for the Pareto gate.
|
|
650
|
-
for (const v of versions) {
|
|
651
|
-
if (!v.promptText)
|
|
652
|
-
continue;
|
|
653
|
-
const metrics = await this.tracker.getAverageMetrics(agentName, v.version);
|
|
654
|
-
if (Object.keys(metrics).length === 0)
|
|
655
|
-
continue; // no runs for this version
|
|
656
|
-
scored.push({ id: v.version, prompt: v.promptText, metrics });
|
|
657
|
-
}
|
|
756
|
+
const scored = await this.buildScoredHistory(agentName);
|
|
658
757
|
if (scored.length < 2)
|
|
659
758
|
return null; // need two parents to merge
|
|
660
759
|
// Two best Pareto-front members (distinct versions, scalarised tie-break).
|