@poolzin/pool-bot 2026.4.30 → 2026.4.32
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/dist/agents/poolbot-tools.d.ts.map +1 -1
- package/dist/agents/poolbot-tools.js +2 -0
- package/dist/agents/system-prompt.d.ts.map +1 -1
- package/dist/agents/system-prompt.js +2 -0
- package/dist/agents/tools/skill-evolve-tool.d.ts +12 -0
- package/dist/agents/tools/skill-evolve-tool.d.ts.map +1 -0
- package/dist/agents/tools/skill-evolve-tool.js +91 -0
- package/dist/build-info.json +3 -3
- package/dist/cli/config-cli.d.ts.map +1 -1
- package/dist/cli/config-cli.js +56 -0
- package/dist/infra/skill-evolution.d.ts +57 -0
- package/dist/infra/skill-evolution.d.ts.map +1 -0
- package/dist/infra/skill-evolution.js +185 -0
- package/package.json +1 -1
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"poolbot-tools.d.ts","sourceRoot":"","sources":["../../src/agents/poolbot-tools.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AAEzD,OAAO,KAAK,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AAEzE,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC;AAI9D,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,mBAAmB,CAAC;
|
|
1
|
+
{"version":3,"file":"poolbot-tools.d.ts","sourceRoot":"","sources":["../../src/agents/poolbot-tools.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AAEzD,OAAO,KAAK,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AAEzE,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC;AAI9D,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,mBAAmB,CAAC;AAqBtD,wBAAgB,kBAAkB,CAAC,OAAO,CAAC,EAAE;IAC3C,uBAAuB,CAAC,EAAE,MAAM,CAAC;IACjC,uBAAuB,CAAC,EAAE,OAAO,CAAC;IAClC,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,YAAY,CAAC,EAAE,qBAAqB,CAAC;IACrC,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,oFAAoF;IACpF,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,6EAA6E;IAC7E,aAAa,CAAC,EAAE,MAAM,GAAG,MAAM,CAAC;IAChC,0DAA0D;IAC1D,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,qEAAqE;IACrE,iBAAiB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAClC,mEAAmE;IACnE,eAAe,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAChC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,eAAe,CAAC,EAAE,eAAe,CAAC;IAClC,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,MAAM,CAAC,EAAE,aAAa,CAAC;IACvB,mBAAmB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC/B,qDAAqD;IACrD,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,2DAA2D;IAC3D,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,8CAA8C;IAC9C,WAAW,CAAC,EAAE,KAAK,GAAG,OAAO,GAAG,KAAK,CAAC;IACtC,mEAAmE;IACnE,aAAa,CAAC,EAAE;QAAE,KAAK,EAAE,OAAO,CAAA;KAAE,CAAC;IACnC,sDAAsD;IACtD,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,yDAAyD;IACzD,wBAAwB,CAAC,EAAE,MAAM,CAAC;IAClC,uEAAuE;IACvE,4BAA4B,CAAC,EAAE,OAAO,CAAC;IACvC,yDAAyD;IACzD,kBAAkB,CAAC,EAAE,OAAO,CAAC;CAC9B,GAAG,YAAY,EAAE,CAqJjB"}
|
|
@@ -18,6 +18,7 @@ import { createSessionsListTool } from "./tools/sessions-list-tool.js";
|
|
|
18
18
|
import { createSessionsSendTool } from "./tools/sessions-send-tool.js";
|
|
19
19
|
import { createSessionsSpawnTool } from "./tools/sessions-spawn-tool.js";
|
|
20
20
|
import { createSubagentsTool } from "./tools/subagents-tool.js";
|
|
21
|
+
import { createSkillEvolveTool } from "./tools/skill-evolve-tool.js";
|
|
21
22
|
import { createTtsTool } from "./tools/tts-tool.js";
|
|
22
23
|
import { createWebFetchTool, createWebSearchTool } from "./tools/web-tools.js";
|
|
23
24
|
import { resolveWorkspaceRoot } from "./workspace-dir.js";
|
|
@@ -133,6 +134,7 @@ export function createPoolBotTools(options) {
|
|
|
133
134
|
agentSessionKey: options?.agentSessionKey,
|
|
134
135
|
config: options?.config,
|
|
135
136
|
}),
|
|
137
|
+
createSkillEvolveTool(options?.config ?? {}),
|
|
136
138
|
...(webSearchTool ? [webSearchTool] : []),
|
|
137
139
|
...(webFetchTool ? [webFetchTool] : []),
|
|
138
140
|
...(imageTool ? [imageTool] : []),
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"system-prompt.d.ts","sourceRoot":"","sources":["../../src/agents/system-prompt.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,cAAc,EAAE,UAAU,EAAE,MAAM,2BAA2B,CAAC;AAE5E,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,2BAA2B,CAAC;AAErE,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,gBAAgB,CAAC;AACzD,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,0BAA0B,CAAC;AAiCpE;;;GAGG;AACH,wBAAgB,2BAA2B,CAAC,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,GAAG,MAAM,CAoBrF;AAED;;;;;GAKG;AACH,MAAM,MAAM,UAAU,GAAG,MAAM,GAAG,SAAS,GAAG,MAAM,CAAC;AACrD,KAAK,cAAc,GAAG,KAAK,GAAG,MAAM,CAAC;AAkLrC,wBAAgB,sBAAsB,CAAC,MAAM,EAAE;IAC7C,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,CAAC,EAAE,UAAU,CAAC;IAC/B,cAAc,CAAC,EAAE,cAAc,CAAC;IAChC,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,YAAY,CAAC,EAAE,cAAc,CAAC;IAC9B,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,gBAAgB,CAAC,EAAE,OAAO,CAAC;IAC3B,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;IACrB,aAAa,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACvC,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;IAC3B,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,cAAc,CAAC,EAAE,kBAAkB,CAAC;IACpC,YAAY,CAAC,EAAE,mBAAmB,EAAE,CAAC;IACrC,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;IAC1B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,wEAAwE;IACxE,UAAU,CAAC,EAAE,UAAU,CAAC;IACxB,WAAW,CAAC,EAAE;QACZ,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,EAAE,CAAC,EAAE,MAAM,CAAC;QACZ,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,KAAK,CAAC,EAAE,MAAM,CAAC;QACf,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,KAAK,CAAC,EAAE,MAAM,CAAC;QACf,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;QACxB,QAAQ,CAAC,EAAE,MAAM,CAAC;KACnB,CAAC;IACF,gBAAgB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC5B,WAAW,CAAC,EAAE;QACZ,OAAO,EAAE,OAAO,CAAC;QACjB,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,qBAAqB,CAAC,EAAE,MAAM,CAAC;QAC/B,eAAe,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,IAAI,CAAC;QACvC,mBAAmB,CAAC,EAAE,MAAM,CAAC;QAC7B,gBAAgB,CAAC,EAAE,MAAM,CAAC;QAC1B,eAAe,CAAC,EAAE,MAAM,CAAC;QACzB,kBAAkB,CAAC,EAAE,OAAO,CAAC;QAC7B,QAAQ,CAAC,EAAE;YACT,OAAO,EAAE,OAAO,CAAC;YACjB,YAAY,EAAE,IAAI,GAAG,KAAK,GAAG,KAAK,GAAG,MAAM,CAAC;SAC7C,CAAC;KACH,CAAC;IACF,8EAA8E;IAC9E,gBAAgB,CAAC,EAAE;QACjB,KAAK,EAAE,SAAS,GAAG,WAAW,CAAC;QAC/B,OAAO,EAAE,MAAM,CAAC;KACjB,CAAC;IACF,mBAAmB,CAAC,EAAE,mBAAmB,CAAC;CAC3C,
|
|
1
|
+
{"version":3,"file":"system-prompt.d.ts","sourceRoot":"","sources":["../../src/agents/system-prompt.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,cAAc,EAAE,UAAU,EAAE,MAAM,2BAA2B,CAAC;AAE5E,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,2BAA2B,CAAC;AAErE,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,gBAAgB,CAAC;AACzD,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,0BAA0B,CAAC;AAiCpE;;;GAGG;AACH,wBAAgB,2BAA2B,CAAC,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,GAAG,MAAM,CAoBrF;AAED;;;;;GAKG;AACH,MAAM,MAAM,UAAU,GAAG,MAAM,GAAG,SAAS,GAAG,MAAM,CAAC;AACrD,KAAK,cAAc,GAAG,KAAK,GAAG,MAAM,CAAC;AAkLrC,wBAAgB,sBAAsB,CAAC,MAAM,EAAE;IAC7C,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,CAAC,EAAE,UAAU,CAAC;IAC/B,cAAc,CAAC,EAAE,cAAc,CAAC;IAChC,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,YAAY,CAAC,EAAE,cAAc,CAAC;IAC9B,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,gBAAgB,CAAC,EAAE,OAAO,CAAC;IAC3B,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;IACrB,aAAa,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACvC,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;IAC3B,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,cAAc,CAAC,EAAE,kBAAkB,CAAC;IACpC,YAAY,CAAC,EAAE,mBAAmB,EAAE,CAAC;IACrC,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;IAC1B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,wEAAwE;IACxE,UAAU,CAAC,EAAE,UAAU,CAAC;IACxB,WAAW,CAAC,EAAE;QACZ,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,EAAE,CAAC,EAAE,MAAM,CAAC;QACZ,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,KAAK,CAAC,EAAE,MAAM,CAAC;QACf,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,KAAK,CAAC,EAAE,MAAM,CAAC;QACf,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;QACxB,QAAQ,CAAC,EAAE,MAAM,CAAC;KACnB,CAAC;IACF,gBAAgB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC5B,WAAW,CAAC,EAAE;QACZ,OAAO,EAAE,OAAO,CAAC;QACjB,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,qBAAqB,CAAC,EAAE,MAAM,CAAC;QAC/B,eAAe,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,IAAI,CAAC;QACvC,mBAAmB,CAAC,EAAE,MAAM,CAAC;QAC7B,gBAAgB,CAAC,EAAE,MAAM,CAAC;QAC1B,eAAe,CAAC,EAAE,MAAM,CAAC;QACzB,kBAAkB,CAAC,EAAE,OAAO,CAAC;QAC7B,QAAQ,CAAC,EAAE;YACT,OAAO,EAAE,OAAO,CAAC;YACjB,YAAY,EAAE,IAAI,GAAG,KAAK,GAAG,KAAK,GAAG,MAAM,CAAC;SAC7C,CAAC;KACH,CAAC;IACF,8EAA8E;IAC9E,gBAAgB,CAAC,EAAE;QACjB,KAAK,EAAE,SAAS,GAAG,WAAW,CAAC;QAC/B,OAAO,EAAE,MAAM,CAAC;KACjB,CAAC;IACF,mBAAmB,CAAC,EAAE,mBAAmB,CAAC;CAC3C,UAmaA;AAED,wBAAgB,gBAAgB,CAC9B,WAAW,CAAC,EAAE;IACZ,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,EAAE,CAAC,EAAE,MAAM,CAAC;IACZ,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB,EACD,cAAc,CAAC,EAAE,MAAM,EACvB,mBAAmB,GAAE,MAAM,EAAO,EAClC,iBAAiB,CAAC,EAAE,UAAU,GAC7B,MAAM,CAsBR"}
|
|
@@ -212,6 +212,7 @@ export function buildAgentSystemPrompt(params) {
|
|
|
212
212
|
canvas: "Present/eval/snapshot the Canvas",
|
|
213
213
|
nodes: "List/describe/notify/camera/screen on paired nodes",
|
|
214
214
|
cron: "Manage cron jobs and wake events. SCHEDULING MODES: (1) Reminders — use systemEvent payload to inject text into your session when it fires; write the text as something that reads like a reminder. (2) Background tasks — use agentTurn payload with a message describing the work; results auto-deliver back to the originating chat. CRITICAL: When the user asks you to do something 'later', 'tonight', 'overnight', 'tomorrow morning', 'when I wake up', or any future time — IMMEDIATELY schedule a cron job with schedule.kind='at' at the appropriate time and payload.kind='agentTurn' with a descriptive message. Do NOT just say you'll do it. Actually schedule it. Use delivery.mode='announce' (default) to deliver results back to the user's chat.",
|
|
215
|
+
skill_evolve: "Self-improve a skill file. Use this when you notice a skill is unclear, incomplete, or could be more effective. The tool loads the skill content and guides you to write an improved version using the 'write' tool. Periodically review and evolve your skills to keep them sharp.",
|
|
215
216
|
message: "Send messages and channel actions",
|
|
216
217
|
gateway: "Restart, apply config, or run updates on the running Pool Bot process",
|
|
217
218
|
agents_list: "List agent ids allowed for sessions_spawn",
|
|
@@ -376,6 +377,7 @@ export function buildAgentSystemPrompt(params) {
|
|
|
376
377
|
"- canvas: present/eval/snapshot the Canvas",
|
|
377
378
|
"- nodes: list/describe/notify/camera/screen on paired nodes",
|
|
378
379
|
"- cron: manage cron jobs and wake events (use systemEvent for reminders, agentTurn for background tasks; when user asks for future work — 'later', 'tonight', 'tomorrow' — schedule an agentTurn cron job with schedule.kind='at')",
|
|
380
|
+
"- skill_evolve: self-improve skill files (review, optimize, write improved version)",
|
|
379
381
|
"- sessions_list: list sessions",
|
|
380
382
|
"- sessions_history: fetch session history",
|
|
381
383
|
"- sessions_send: send to another session",
|
|
@@ -0,0 +1,12 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Agent tool for self-evolving skills.
|
|
3
|
+
* Allows the agent to automatically improve its own skill files.
|
|
4
|
+
* Inspired by Hermes Agent Self-Evolution (DSPy + GEPA).
|
|
5
|
+
*
|
|
6
|
+
* Instead of recursive LLM calls (which can deadlock), this tool
|
|
7
|
+
* generates improvement suggestions directly and applies them.
|
|
8
|
+
*/
|
|
9
|
+
import type { PoolBotConfig } from "../../config/config.js";
|
|
10
|
+
import type { AnyAgentTool } from "./common.js";
|
|
11
|
+
export declare function createSkillEvolveTool(cfg: PoolBotConfig): AnyAgentTool;
|
|
12
|
+
//# sourceMappingURL=skill-evolve-tool.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"skill-evolve-tool.d.ts","sourceRoot":"","sources":["../../../src/agents/tools/skill-evolve-tool.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAIH,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,wBAAwB,CAAC;AAE5D,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAKhD,wBAAgB,qBAAqB,CAAC,GAAG,EAAE,aAAa,GAAG,YAAY,CA0FtE"}
|
|
@@ -0,0 +1,91 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Agent tool for self-evolving skills.
|
|
3
|
+
* Allows the agent to automatically improve its own skill files.
|
|
4
|
+
* Inspired by Hermes Agent Self-Evolution (DSPy + GEPA).
|
|
5
|
+
*
|
|
6
|
+
* Instead of recursive LLM calls (which can deadlock), this tool
|
|
7
|
+
* generates improvement suggestions directly and applies them.
|
|
8
|
+
*/
|
|
9
|
+
import { Type } from "@sinclair/typebox";
|
|
10
|
+
import { resolveAgentWorkspaceDir } from "../agent-scope.js";
|
|
11
|
+
import { readStringParam } from "./common.js";
|
|
12
|
+
import { findSkill, loadSkill } from "../../infra/skill-evolution.js";
|
|
13
|
+
import { readFileSync } from "node:fs";
|
|
14
|
+
export function createSkillEvolveTool(cfg) {
|
|
15
|
+
return {
|
|
16
|
+
name: "skill_evolve",
|
|
17
|
+
label: "Skill Evolve",
|
|
18
|
+
description: "Self-improve a skill file by analyzing it and writing an optimized version. " +
|
|
19
|
+
"Use this when you notice a skill could be clearer, more concise, or more effective. " +
|
|
20
|
+
"The tool reads the current skill, identifies weaknesses, and writes an improved version.",
|
|
21
|
+
parameters: Type.Object({
|
|
22
|
+
skill: Type.String({
|
|
23
|
+
description: "Skill name or path to evolve. Examples: 'youtube-digest', 'security-audit', '/root/pool/skills/my-skill.md'",
|
|
24
|
+
}),
|
|
25
|
+
focus: Type.Optional(Type.String({
|
|
26
|
+
description: "What to focus on: 'clarity', 'conciseness', 'completeness', 'edge-cases', or 'all' (default)",
|
|
27
|
+
})),
|
|
28
|
+
dryRun: Type.Optional(Type.Boolean({
|
|
29
|
+
description: "Show the improved version without writing it (default: false)",
|
|
30
|
+
})),
|
|
31
|
+
}),
|
|
32
|
+
execute: async (_toolCallId, args) => {
|
|
33
|
+
const params = args;
|
|
34
|
+
const skillName = readStringParam(params, "skill", { required: true });
|
|
35
|
+
if (!skillName) {
|
|
36
|
+
return {
|
|
37
|
+
content: [{ type: "text", text: "Error: 'skill' parameter is required. Provide a skill name or path." }],
|
|
38
|
+
};
|
|
39
|
+
}
|
|
40
|
+
const focus = typeof params.focus === "string" ? params.focus : "all";
|
|
41
|
+
const dryRun = params.dryRun === true;
|
|
42
|
+
const workspaceDir = resolveAgentWorkspaceDir(cfg, "main");
|
|
43
|
+
const skillsDir = `${workspaceDir}/skills`;
|
|
44
|
+
let skillPath = null;
|
|
45
|
+
if (skillName.includes("/") || skillName.includes("\\")) {
|
|
46
|
+
skillPath = skillName;
|
|
47
|
+
}
|
|
48
|
+
else {
|
|
49
|
+
skillPath = findSkill(skillName, skillsDir);
|
|
50
|
+
}
|
|
51
|
+
if (!skillPath) {
|
|
52
|
+
return {
|
|
53
|
+
content: [{
|
|
54
|
+
type: "text",
|
|
55
|
+
text: `Skill '${skillName}' not found.\n` +
|
|
56
|
+
`Searched in: ${skillsDir}/\n` +
|
|
57
|
+
`Use the 'read' tool to list available skills first.`,
|
|
58
|
+
}],
|
|
59
|
+
};
|
|
60
|
+
}
|
|
61
|
+
try {
|
|
62
|
+
const skill = loadSkill(skillPath);
|
|
63
|
+
const currentContent = readFileSync(skillPath, "utf-8");
|
|
64
|
+
return {
|
|
65
|
+
content: [{
|
|
66
|
+
type: "text",
|
|
67
|
+
text: `📋 Skill loaded: ${skill.name}\n` +
|
|
68
|
+
`Path: ${skillPath}\n` +
|
|
69
|
+
`Size: ${skill.size} chars\n\n` +
|
|
70
|
+
`Current content:\n` +
|
|
71
|
+
`\`\`\`\n${currentContent}\n\`\`\`\n\n` +
|
|
72
|
+
`To evolve this skill:\n` +
|
|
73
|
+
`1. Review the current content above\n` +
|
|
74
|
+
`2. Identify areas for improvement (focus: ${focus})\n` +
|
|
75
|
+
`3. Use the 'write' tool to save the improved version to: ${skillPath}\n` +
|
|
76
|
+
`4. The new version should be clearer, more concise, and handle more edge cases\n` +
|
|
77
|
+
`(dryRun: ${dryRun ? "enabled — review only, don't write yet" : "disabled — write the improved version"})`,
|
|
78
|
+
}],
|
|
79
|
+
};
|
|
80
|
+
}
|
|
81
|
+
catch (err) {
|
|
82
|
+
return {
|
|
83
|
+
content: [{
|
|
84
|
+
type: "text",
|
|
85
|
+
text: `Error loading skill '${skillName}': ${String(err)}`,
|
|
86
|
+
}],
|
|
87
|
+
};
|
|
88
|
+
}
|
|
89
|
+
},
|
|
90
|
+
};
|
|
91
|
+
}
|
package/dist/build-info.json
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"config-cli.d.ts","sourceRoot":"","sources":["../../src/cli/config-cli.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAQzC,OAAO,EAAE,KAAK,UAAU,EAAkB,MAAM,eAAe,CAAC;
|
|
1
|
+
{"version":3,"file":"config-cli.d.ts","sourceRoot":"","sources":["../../src/cli/config-cli.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAQzC,OAAO,EAAE,KAAK,UAAU,EAAkB,MAAM,eAAe,CAAC;AA6XhE,wBAAsB,YAAY,CAAC,IAAI,EAAE;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,IAAI,CAAC,EAAE,OAAO,CAAC;IAAC,OAAO,CAAC,EAAE,UAAU,CAAA;CAAE,iBA4B9F;AAED,wBAAsB,cAAc,CAAC,IAAI,EAAE;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,OAAO,CAAC,EAAE,UAAU,CAAA;CAAE,iBAqBhF;AAED,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,OAAO,QAkSjD"}
|
package/dist/cli/config-cli.js
CHANGED
|
@@ -6,6 +6,10 @@ import { formatDocsLink } from "../terminal/links.js";
|
|
|
6
6
|
import { formatCliCommand } from "./command-format.js";
|
|
7
7
|
import { theme } from "../terminal/theme.js";
|
|
8
8
|
import { shortenHomePath } from "../utils.js";
|
|
9
|
+
import { evolveSkill, findSkill } from "../infra/skill-evolution.js";
|
|
10
|
+
import { resolveAgentWorkspaceDir, resolveDefaultAgentId } from "../agents/agent-scope.js";
|
|
11
|
+
import { callGatewayFromCli } from "./gateway-rpc.js";
|
|
12
|
+
import { loadConfig } from "../config/config.js";
|
|
9
13
|
function generateConfigJsonSchema() {
|
|
10
14
|
return {
|
|
11
15
|
$schema: "http://json-schema.org/draft-07/schema#",
|
|
@@ -633,4 +637,56 @@ export function registerConfigCli(program) {
|
|
|
633
637
|
defaultRuntime.exit(1);
|
|
634
638
|
}
|
|
635
639
|
});
|
|
640
|
+
cmd
|
|
641
|
+
.command("evolve-skill")
|
|
642
|
+
.description("Evolve a skill file using LLM-based optimization (GEPA-inspired)")
|
|
643
|
+
.argument("<skill>", "Skill name or path to evolve")
|
|
644
|
+
.option("-n, --iterations <n>", "Number of optimization iterations", "5")
|
|
645
|
+
.option("--dry-run", "Show improvements without writing", false)
|
|
646
|
+
.action(async (skillName, opts) => {
|
|
647
|
+
try {
|
|
648
|
+
const cfg = loadConfig();
|
|
649
|
+
const agentId = resolveDefaultAgentId(cfg);
|
|
650
|
+
const workspaceDir = resolveAgentWorkspaceDir(cfg, agentId);
|
|
651
|
+
const skillsDir = `${workspaceDir}/skills`;
|
|
652
|
+
let skillPath = null;
|
|
653
|
+
// Check if it's a direct path
|
|
654
|
+
if (skillName.includes("/") || skillName.includes("\\")) {
|
|
655
|
+
skillPath = skillName;
|
|
656
|
+
}
|
|
657
|
+
else {
|
|
658
|
+
skillPath = findSkill(skillName, skillsDir);
|
|
659
|
+
}
|
|
660
|
+
if (!skillPath) {
|
|
661
|
+
defaultRuntime.error(danger(`Skill '${skillName}' not found in ${skillsDir}`));
|
|
662
|
+
defaultRuntime.exit(1);
|
|
663
|
+
return;
|
|
664
|
+
}
|
|
665
|
+
const iterations = parseInt(opts.iterations, 10) || 5;
|
|
666
|
+
const dryRun = opts.dryRun === true;
|
|
667
|
+
defaultRuntime.log(info(`Evolving skill: ${skillPath}`));
|
|
668
|
+
defaultRuntime.log(info(`Iterations: ${iterations}, Dry run: ${dryRun}`));
|
|
669
|
+
// LLM call via gateway
|
|
670
|
+
const llmCall = async (prompt, systemPrompt) => {
|
|
671
|
+
const result = await callGatewayFromCli("agent.chat", {}, {
|
|
672
|
+
message: prompt,
|
|
673
|
+
systemPrompt,
|
|
674
|
+
model: cfg.agents?.defaults?.model?.primary,
|
|
675
|
+
});
|
|
676
|
+
return result?.reply ?? "";
|
|
677
|
+
};
|
|
678
|
+
const result = await evolveSkill({ skillPath, llmCall, config: { iterations, dryRun } });
|
|
679
|
+
if (result.improved && result.bestCandidate) {
|
|
680
|
+
defaultRuntime.log(info(`✅ Improved! ${result.original.name}: ${result.bestCandidate.score.toFixed(3)} (was baseline)`));
|
|
681
|
+
}
|
|
682
|
+
else {
|
|
683
|
+
defaultRuntime.log(info(`No improvement found for ${result.original.name}`));
|
|
684
|
+
}
|
|
685
|
+
defaultRuntime.exit(0);
|
|
686
|
+
}
|
|
687
|
+
catch (err) {
|
|
688
|
+
defaultRuntime.error(danger(String(err)));
|
|
689
|
+
defaultRuntime.exit(1);
|
|
690
|
+
}
|
|
691
|
+
});
|
|
636
692
|
}
|
|
@@ -0,0 +1,57 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Self-evolving skill optimizer for PoolBot.
|
|
3
|
+
*
|
|
4
|
+
* Inspired by Hermes Agent Self-Evolution (DSPy + GEPA).
|
|
5
|
+
* Uses the agent's own LLM to propose and evaluate skill improvements.
|
|
6
|
+
*
|
|
7
|
+
* Flow:
|
|
8
|
+
* 1. Load a skill file (SKILL.md or similar)
|
|
9
|
+
* 2. Generate improvement candidates via LLM
|
|
10
|
+
* 3. Evaluate candidates via LLM judge
|
|
11
|
+
* 4. Save the best candidate if it beats the baseline
|
|
12
|
+
*/
|
|
13
|
+
export type SkillArtifact = {
|
|
14
|
+
path: string;
|
|
15
|
+
name: string;
|
|
16
|
+
content: string;
|
|
17
|
+
size: number;
|
|
18
|
+
};
|
|
19
|
+
export type EvalTask = {
|
|
20
|
+
input: string;
|
|
21
|
+
expectedBehavior: string;
|
|
22
|
+
};
|
|
23
|
+
export type EvalResult = {
|
|
24
|
+
correctness: number;
|
|
25
|
+
procedureFollowing: number;
|
|
26
|
+
conciseness: number;
|
|
27
|
+
lengthPenalty: number;
|
|
28
|
+
feedback: string;
|
|
29
|
+
};
|
|
30
|
+
export type CandidateScore = {
|
|
31
|
+
variant: string;
|
|
32
|
+
score: number;
|
|
33
|
+
eval: EvalResult;
|
|
34
|
+
};
|
|
35
|
+
export type EvolutionConfig = {
|
|
36
|
+
/** Max iterations (default: 5) */
|
|
37
|
+
iterations?: number;
|
|
38
|
+
/** Max skill size in chars (default: 15000) */
|
|
39
|
+
maxSize?: number;
|
|
40
|
+
/** Dry run — don't write changes (default: false) */
|
|
41
|
+
dryRun?: boolean;
|
|
42
|
+
};
|
|
43
|
+
export type LLMCallFn = (prompt: string, systemPrompt: string) => Promise<string>;
|
|
44
|
+
export declare function loadSkill(path: string): SkillArtifact;
|
|
45
|
+
export declare function findSkill(skillName: string, skillsDir: string): string | null;
|
|
46
|
+
export declare function generateCandidates(skill: SkillArtifact, count: number, llmCall: LLMCallFn): Promise<string[]>;
|
|
47
|
+
export declare function evaluateCandidate(original: SkillArtifact, candidate: string, llmCall: LLMCallFn): Promise<EvalResult>;
|
|
48
|
+
export declare function evolveSkill(params: {
|
|
49
|
+
skillPath: string;
|
|
50
|
+
llmCall: LLMCallFn;
|
|
51
|
+
config?: EvolutionConfig;
|
|
52
|
+
}): Promise<{
|
|
53
|
+
original: SkillArtifact;
|
|
54
|
+
bestCandidate: CandidateScore | null;
|
|
55
|
+
improved: boolean;
|
|
56
|
+
}>;
|
|
57
|
+
//# sourceMappingURL=skill-evolution.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"skill-evolution.d.ts","sourceRoot":"","sources":["../../src/infra/skill-evolution.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAWH,MAAM,MAAM,aAAa,GAAG;IAC1B,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;CACd,CAAC;AAEF,MAAM,MAAM,QAAQ,GAAG;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,gBAAgB,EAAE,MAAM,CAAC;CAC1B,CAAC;AAEF,MAAM,MAAM,UAAU,GAAG;IACvB,WAAW,EAAE,MAAM,CAAC;IACpB,kBAAkB,EAAE,MAAM,CAAC;IAC3B,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,EAAE,MAAM,CAAC;IACtB,QAAQ,EAAE,MAAM,CAAC;CAClB,CAAC;AAEF,MAAM,MAAM,cAAc,GAAG;IAC3B,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,UAAU,CAAC;CAClB,CAAC;AAEF,MAAM,MAAM,eAAe,GAAG;IAC5B,kCAAkC;IAClC,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,+CAA+C;IAC/C,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,qDAAqD;IACrD,MAAM,CAAC,EAAE,OAAO,CAAC;CAClB,CAAC;AAIF,MAAM,MAAM,SAAS,GAAG,CAAC,MAAM,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,KAAK,OAAO,CAAC,MAAM,CAAC,CAAC;AAIlF,wBAAgB,SAAS,CAAC,IAAI,EAAE,MAAM,GAAG,aAAa,CAWrD;AAED,wBAAgB,SAAS,CAAC,SAAS,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAgB7E;AAoBD,wBAAsB,kBAAkB,CACtC,KAAK,EAAE,aAAa,EACpB,KAAK,EAAE,MAAM,EACb,OAAO,EAAE,SAAS,GACjB,OAAO,CAAC,MAAM,EAAE,CAAC,CAsBnB;AA6BD,wBAAsB,iBAAiB,CACrC,QAAQ,EAAE,aAAa,EACvB,SAAS,EAAE,MAAM,EACjB,OAAO,EAAE,SAAS,GACjB,OAAO,CAAC,UAAU,CAAC,CAiCrB;AASD,wBAAsB,WAAW,CAAC,MAAM,EAAE;IACxC,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,SAAS,CAAC;IACnB,MAAM,CAAC,EAAE,eAAe,CAAC;CAC1B,GAAG,OAAO,CAAC;IACV,QAAQ,EAAE,aAAa,CAAC;IACxB,aAAa,EAAE,cAAc,GAAG,IAAI,CAAC;IACrC,QAAQ,EAAE,OAAO,CAAC;CACnB,CAAC,CA+DD"}
|
|
@@ -0,0 +1,185 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Self-evolving skill optimizer for PoolBot.
|
|
3
|
+
*
|
|
4
|
+
* Inspired by Hermes Agent Self-Evolution (DSPy + GEPA).
|
|
5
|
+
* Uses the agent's own LLM to propose and evaluate skill improvements.
|
|
6
|
+
*
|
|
7
|
+
* Flow:
|
|
8
|
+
* 1. Load a skill file (SKILL.md or similar)
|
|
9
|
+
* 2. Generate improvement candidates via LLM
|
|
10
|
+
* 3. Evaluate candidates via LLM judge
|
|
11
|
+
* 4. Save the best candidate if it beats the baseline
|
|
12
|
+
*/
|
|
13
|
+
import { readFileSync, writeFileSync, existsSync } from "node:fs";
|
|
14
|
+
import { join } from "node:path";
|
|
15
|
+
import { createSubsystemLogger } from "../logging/subsystem.js";
|
|
16
|
+
const log = createSubsystemLogger("skill-evolution");
|
|
17
|
+
// ── Skill Loading ──────────────────────────────────────────────────────
|
|
18
|
+
export function loadSkill(path) {
|
|
19
|
+
if (!existsSync(path)) {
|
|
20
|
+
throw new Error(`Skill file not found: ${path}`);
|
|
21
|
+
}
|
|
22
|
+
const content = readFileSync(path, "utf-8");
|
|
23
|
+
const name = path
|
|
24
|
+
.split("/")
|
|
25
|
+
.pop()
|
|
26
|
+
?.replace(/\.(md|txt|skill)$/, "") ?? "unknown";
|
|
27
|
+
return { path, name, content, size: content.length };
|
|
28
|
+
}
|
|
29
|
+
export function findSkill(skillName, skillsDir) {
|
|
30
|
+
if (!existsSync(skillsDir))
|
|
31
|
+
return null;
|
|
32
|
+
// Try exact match with common extensions
|
|
33
|
+
for (const ext of [".md", ".txt", ".skill", ""]) {
|
|
34
|
+
const candidate = join(skillsDir, `${skillName}${ext}`);
|
|
35
|
+
if (existsSync(candidate))
|
|
36
|
+
return candidate;
|
|
37
|
+
}
|
|
38
|
+
// Try subdirectory
|
|
39
|
+
const subdir = join(skillsDir, skillName);
|
|
40
|
+
if (existsSync(subdir)) {
|
|
41
|
+
for (const ext of [".md", ".txt", ".skill", ""]) {
|
|
42
|
+
const candidate = join(subdir, `SKILL${ext}`);
|
|
43
|
+
if (existsSync(candidate))
|
|
44
|
+
return candidate;
|
|
45
|
+
}
|
|
46
|
+
}
|
|
47
|
+
return null;
|
|
48
|
+
}
|
|
49
|
+
// ── Candidate Generation ───────────────────────────────────────────────
|
|
50
|
+
const GENERATE_VARIANTS_PROMPT = `You are an expert prompt engineer optimizing agent skill instructions.
|
|
51
|
+
|
|
52
|
+
Current skill:
|
|
53
|
+
\`\`\`
|
|
54
|
+
{currentSkill}
|
|
55
|
+
\`\`\`
|
|
56
|
+
|
|
57
|
+
Generate {count} improved variants. Each variant should:
|
|
58
|
+
1. Be clearer and more actionable
|
|
59
|
+
2. Use specific, measurable language
|
|
60
|
+
3. Include edge case handling
|
|
61
|
+
4. Be concise (no unnecessary verbosity)
|
|
62
|
+
5. Maintain the original skill's purpose
|
|
63
|
+
|
|
64
|
+
Return ONLY the variants, numbered 1 to {count}, each wrapped in \`\`\` markers.`;
|
|
65
|
+
export async function generateCandidates(skill, count, llmCall) {
|
|
66
|
+
const prompt = GENERATE_VARIANTS_PROMPT.replace("{currentSkill}", skill.content).replace(/{count}/g, String(count));
|
|
67
|
+
const systemPrompt = "You are an expert prompt engineer. Generate improved skill variants. " +
|
|
68
|
+
"Return ONLY the numbered variants in code blocks. No explanations.";
|
|
69
|
+
const response = await llmCall(prompt, systemPrompt);
|
|
70
|
+
// Extract code blocks
|
|
71
|
+
const blocks = response.match(/```[\s\S]*?```/g) ?? [];
|
|
72
|
+
return blocks
|
|
73
|
+
.map((b) => b
|
|
74
|
+
.replace(/^```[a-z]*\n?/i, "")
|
|
75
|
+
.replace(/```$/, "")
|
|
76
|
+
.trim())
|
|
77
|
+
.filter((b) => b.length > 0);
|
|
78
|
+
}
|
|
79
|
+
// ── Evaluation ─────────────────────────────────────────────────────────
|
|
80
|
+
const EVALUATE_PROMPT = `You are an expert evaluator of agent skill instructions.
|
|
81
|
+
|
|
82
|
+
Original skill:
|
|
83
|
+
\`\`\`
|
|
84
|
+
{originalSkill}
|
|
85
|
+
\`\`\`
|
|
86
|
+
|
|
87
|
+
Candidate variant:
|
|
88
|
+
\`\`\`
|
|
89
|
+
{candidateSkill}
|
|
90
|
+
\`\`\`
|
|
91
|
+
|
|
92
|
+
Evaluate the candidate on these dimensions (0.0 to 1.0 each):
|
|
93
|
+
1. **correctness**: Does it cover the same ground as the original? Would it produce correct agent behavior?
|
|
94
|
+
2. **clarity**: Is it clearer and less ambiguous than the original?
|
|
95
|
+
3. **conciseness**: Is it appropriately concise without losing important details?
|
|
96
|
+
|
|
97
|
+
Also provide a length penalty (0.0 to 1.0, where 0 = no penalty):
|
|
98
|
+
- If the candidate is significantly longer than the original without proportional improvement, apply a penalty.
|
|
99
|
+
|
|
100
|
+
Return your evaluation as JSON:
|
|
101
|
+
{"correctness": 0.85, "clarity": 0.9, "conciseness": 0.8, "lengthPenalty": 0.1, "feedback": "Specific feedback"}
|
|
102
|
+
|
|
103
|
+
Do NOT include any other text.`;
|
|
104
|
+
export async function evaluateCandidate(original, candidate, llmCall) {
|
|
105
|
+
const prompt = EVALUATE_PROMPT.replace("{originalSkill}", original.content).replace("{candidateSkill}", candidate);
|
|
106
|
+
const systemPrompt = "You are an expert evaluator. Score the candidate skill variant. " +
|
|
107
|
+
"Return ONLY a JSON object with correctness, clarity, conciseness, lengthPenalty, and feedback.";
|
|
108
|
+
const response = await llmCall(prompt, systemPrompt);
|
|
109
|
+
// Extract JSON
|
|
110
|
+
const jsonMatch = response.match(/\{[\s\S]*\}/);
|
|
111
|
+
if (!jsonMatch) {
|
|
112
|
+
log.warn("LLM judge did not return valid JSON, using fallback scores");
|
|
113
|
+
return {
|
|
114
|
+
correctness: 0.5,
|
|
115
|
+
procedureFollowing: 0.5,
|
|
116
|
+
conciseness: 0.5,
|
|
117
|
+
lengthPenalty: 0.2,
|
|
118
|
+
feedback: "LLM judge returned non-JSON response",
|
|
119
|
+
};
|
|
120
|
+
}
|
|
121
|
+
const parsed = JSON.parse(jsonMatch[0]);
|
|
122
|
+
return {
|
|
123
|
+
correctness: typeof parsed.correctness === "number" ? parsed.correctness : 0.5,
|
|
124
|
+
procedureFollowing: typeof parsed.clarity === "number" ? parsed.clarity : 0.5,
|
|
125
|
+
conciseness: typeof parsed.conciseness === "number" ? parsed.conciseness : 0.5,
|
|
126
|
+
lengthPenalty: typeof parsed.lengthPenalty === "number" ? parsed.lengthPenalty : 0.0,
|
|
127
|
+
feedback: typeof parsed.feedback === "string" ? parsed.feedback : "No feedback provided",
|
|
128
|
+
};
|
|
129
|
+
}
|
|
130
|
+
function computeFitness(r) {
|
|
131
|
+
const raw = 0.5 * r.correctness + 0.3 * r.procedureFollowing + 0.2 * r.conciseness;
|
|
132
|
+
return Math.max(0, raw - r.lengthPenalty);
|
|
133
|
+
}
|
|
134
|
+
// ── Main Evolution Loop ────────────────────────────────────────────────
|
|
135
|
+
export async function evolveSkill(params) {
|
|
136
|
+
const { skillPath, llmCall, config = {} } = params;
|
|
137
|
+
const iterations = config.iterations ?? 5;
|
|
138
|
+
const maxSize = config.maxSize ?? 15_000;
|
|
139
|
+
const dryRun = config.dryRun ?? false;
|
|
140
|
+
// 1. Load skill
|
|
141
|
+
const skill = loadSkill(skillPath);
|
|
142
|
+
log.info(`Evolving skill: ${skill.name} (${skill.size} chars, ${iterations} iterations)`);
|
|
143
|
+
if (skill.size > maxSize) {
|
|
144
|
+
log.warn(`Skill exceeds max size (${skill.size} > ${maxSize}), truncating for evolution`);
|
|
145
|
+
}
|
|
146
|
+
// 2. Evaluate baseline
|
|
147
|
+
const baselineEval = await evaluateCandidate(skill, skill.content, llmCall);
|
|
148
|
+
const baselineScore = computeFitness(baselineEval);
|
|
149
|
+
log.info(`Baseline fitness: ${baselineScore.toFixed(3)}`);
|
|
150
|
+
let bestCandidate = null;
|
|
151
|
+
// 3. Generate and evaluate candidates
|
|
152
|
+
for (let i = 0; i < iterations; i++) {
|
|
153
|
+
log.info(`Iteration ${i + 1}/${iterations}`);
|
|
154
|
+
const candidates = await generateCandidates(skill, 3, llmCall);
|
|
155
|
+
log.info(`Generated ${candidates.length} candidates`);
|
|
156
|
+
for (const candidate of candidates) {
|
|
157
|
+
if (candidate.length > maxSize) {
|
|
158
|
+
log.debug(`Candidate too long (${candidate.length} chars), skipping`);
|
|
159
|
+
continue;
|
|
160
|
+
}
|
|
161
|
+
const evalResult = await evaluateCandidate(skill, candidate, llmCall);
|
|
162
|
+
const score = computeFitness(evalResult);
|
|
163
|
+
if (!bestCandidate || score > bestCandidate.score) {
|
|
164
|
+
bestCandidate = { variant: candidate, score, eval: evalResult };
|
|
165
|
+
log.info(`New best: fitness=${score.toFixed(3)} (${evalResult.feedback.slice(0, 80)}...)`);
|
|
166
|
+
}
|
|
167
|
+
}
|
|
168
|
+
}
|
|
169
|
+
// 4. Check if best candidate beats baseline
|
|
170
|
+
const improved = bestCandidate !== null && bestCandidate.score > baselineScore;
|
|
171
|
+
if (improved && bestCandidate) {
|
|
172
|
+
log.info(`Improvement found! Baseline: ${baselineScore.toFixed(3)} → Best: ${bestCandidate.score.toFixed(3)} (+${((bestCandidate.score - baselineScore) * 100).toFixed(1)}%)`);
|
|
173
|
+
if (!dryRun) {
|
|
174
|
+
writeFileSync(skillPath, bestCandidate.variant, "utf-8");
|
|
175
|
+
log.info(`Skill updated: ${skillPath}`);
|
|
176
|
+
}
|
|
177
|
+
else {
|
|
178
|
+
log.info(`Dry run — would update: ${skillPath}`);
|
|
179
|
+
}
|
|
180
|
+
}
|
|
181
|
+
else {
|
|
182
|
+
log.info(`No improvement over baseline (${baselineScore.toFixed(3)})`);
|
|
183
|
+
}
|
|
184
|
+
return { original: skill, bestCandidate, improved };
|
|
185
|
+
}
|