Self-learning is excluded from the community edition (ADR-0004); the project owner reworks it separately. Cost tracking stays. - agent: delete behaviour evolution (evolve-schema, iterative GEPA, compose-evolution, evolution gates, orchestrator and deploy), improvement and correction detection, AgentLearning, the eval-dataset builder, optimization-capture and the agent_insights tool. Secret detection moves from eval-dataset to secret-patterns. The LLM wiring keeps only the Haiku adapter and judge call used by skill audit and hygiene, so @ax-llm/ax is now declared by agent instead of server. - server: delete /api/evolution/*, the evolution and optimizer services, GEPA expansion of the first chat message, the prompt_optimization and monthly_assessment jobs, the idle-session reviewer and its session-reviewer persona, prompt injection of past corrections and thumbs-down recorded as correction signals. Existing job rows of the removed types are ignored, not deleted. - core: delete OptimizationLogStore; stop re-exporting the improvement signal and evolution run stores. - web: delete the Evolution screen and the two job types in the automation editor. - delete packages/optimizer. Also logs UF-027 (self-learning loops read data nothing collects) and UF-028 (the Settings prompt-shape choice is ignored by chat).
118 lines
4.1 KiB
TypeScript
118 lines
4.1 KiB
TypeScript
/**
|
|
* Skills 2.0 gap A — auto-extract skills from session history.
|
|
*
|
|
* Closes the loop between detectWorkflowPattern (session analysis) and
|
|
* generateSkillMarkdown (SKILL.md file generation). Previously these two
|
|
* primitives existed but were never wired together — SKILL.md files only
|
|
* got created via the explicit create_skill agent tool.
|
|
*
|
|
* This module provides a deterministic pipeline (no LLM required) that:
|
|
* 1. Runs detectWorkflowPattern on a message history
|
|
* 2. Generates a SKILL.md via generateSkillMarkdown
|
|
* 3. Writes it to the personal scope directory (scope: personal frontmatter)
|
|
* 4. Dedups against existing skill files by filename
|
|
*
|
|
* A v2 enhancement can insert an LLM prose-improvement step between
|
|
* detectWorkflowPattern and generateSkillMarkdown without changing this
|
|
* module's public surface.
|
|
*/
|
|
import * as fs from 'node:fs';
|
|
import * as path from 'node:path';
|
|
import {
|
|
detectWorkflowPattern,
|
|
generateSkillMarkdown,
|
|
type SkillTemplate,
|
|
} from './skill-creator.js';
|
|
import { redactSkillContent } from './skill-redaction.js';
|
|
|
|
export interface AutoExtractMessage {
|
|
role: string;
|
|
content: string;
|
|
toolsUsed?: string[];
|
|
}
|
|
|
|
export interface AutoExtractDeps {
|
|
/** Directory to write personal-scope SKILL.md files into */
|
|
personalSkillsDir: string;
|
|
/**
|
|
* Optional LLM enrichment hook. When provided, the template is passed
|
|
* through before serialization so the SKILL.md prose can be rewritten.
|
|
* Must return a SkillTemplate (may be the same object). If the hook
|
|
* throws, the deterministic template is used as a safe fallback.
|
|
*/
|
|
enrichTemplate?: (template: SkillTemplate, messages: AutoExtractMessage[]) => Promise<SkillTemplate>;
|
|
}
|
|
|
|
export interface AutoExtractResult {
|
|
/** Path to the SKILL.md written, or null when nothing was created */
|
|
filePath: string | null;
|
|
/** Why the extraction returned null (when applicable) */
|
|
reason?: 'no_pattern' | 'already_exists';
|
|
/** The template used (present when filePath is non-null) */
|
|
template?: SkillTemplate;
|
|
}
|
|
|
|
/** Produce a kebab-case filename stem from a SkillTemplate name. */
|
|
export function skillFilename(template: SkillTemplate): string {
|
|
return template.name
|
|
.toLowerCase()
|
|
.replace(/[^a-z0-9]+/g, '-')
|
|
.replace(/^-+|-+$/g, '');
|
|
}
|
|
|
|
/**
|
|
* Inject scope + auto-extracted provenance into the generated SKILL.md so
|
|
* gap-E promotion gates recognize this as a personal-scope skill with a
|
|
* clear origin marker. Operates on the frontmatter block only.
|
|
*/
|
|
function annotateWithScope(md: string): string {
|
|
// Insert the scope line right after the description line in the
|
|
// frontmatter block. detectWorkflowPattern + generateSkillMarkdown never
|
|
// emit scope, so no existing line is overwritten.
|
|
return md.replace(
|
|
/(description: [^\n]+)\n/,
|
|
'$1\nscope: personal\n',
|
|
);
|
|
}
|
|
|
|
/**
|
|
* Run the full pipeline end-to-end. Returns null.filePath when no pattern
|
|
* was detected or a skill with the same filename already exists.
|
|
*/
|
|
export async function autoExtractAndCreateSkill(
|
|
messages: AutoExtractMessage[],
|
|
deps: AutoExtractDeps,
|
|
): Promise<AutoExtractResult> {
|
|
const template = detectWorkflowPattern(messages);
|
|
if (!template) {
|
|
return { filePath: null, reason: 'no_pattern' };
|
|
}
|
|
|
|
const stem = skillFilename(template);
|
|
if (!fs.existsSync(deps.personalSkillsDir)) {
|
|
fs.mkdirSync(deps.personalSkillsDir, { recursive: true });
|
|
}
|
|
const filePath = path.join(deps.personalSkillsDir, `${stem}.md`);
|
|
if (fs.existsSync(filePath)) {
|
|
return { filePath: null, reason: 'already_exists', template };
|
|
}
|
|
|
|
// Optional LLM enrichment — isolated in try/catch so a flaky model
|
|
// never fails the deterministic fallback path.
|
|
let finalTemplate = template;
|
|
if (deps.enrichTemplate) {
|
|
try {
|
|
finalTemplate = await deps.enrichTemplate(template, messages);
|
|
} catch {
|
|
finalTemplate = template;
|
|
}
|
|
}
|
|
|
|
// Strip secrets + user-home paths — the auto-extracted pattern is derived
|
|
// from session messages, which can carry both.
|
|
const md = redactSkillContent(annotateWithScope(generateSkillMarkdown(finalTemplate))).content;
|
|
fs.writeFileSync(filePath, md, 'utf-8');
|
|
|
|
return { filePath, template: finalTemplate };
|
|
}
|