refactor: remove self-learning and the optimizer package (P1.4)

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).
This commit is contained in:
shumeiko.it
2026-10-20 00:54:58 +02:00
parent b734c1d3d2
commit 2e1999a3f6
112 changed files with 336 additions and 16165 deletions

View File

@@ -94,7 +94,7 @@ function summarizeInput(input: Record<string, unknown>): string {
* approve blind: the approver sees the skill's name up front and can expand the
* exact bytes `writeSkill` will persist before allowing it. `summarizeInput`
* surfaces none of `{name, content}`, so without this the card showed only the
* tool name — a review-before-apply gap for the self-evolution loop.
* tool name — a review-before-apply gap for agent-authored skills.
*/
function SkillPreview({ name, content }: { name: string; content: string }) {
const [expanded, setExpanded] = useState(false);

View File

@@ -1,6 +1,6 @@
import { useState } from 'react';
import { motion, useReducedMotion } from 'framer-motion';
import { Brain, Clock, Network, Download, Activity, BookOpen, Sparkles, User, Briefcase, ShieldCheck, SlidersHorizontal, ChevronDown } from 'lucide-react';
import { Brain, Clock, Network, Download, Activity, BookOpen, User, Briefcase, ShieldCheck, SlidersHorizontal, ChevronDown } from 'lucide-react';
import type { KGNode, KGEdge } from '@/lib/types';
import { HintTooltip } from '@/components/ui/hint-tooltip';
import { SPRING } from '@/lib/motion/tokens';
@@ -10,7 +10,6 @@ import KnowledgeGraphViewer from './memory/KnowledgeGraphViewer';
import HarvestTab from './memory/HarvestTab';
import WeaverPanel from './memory/WeaverPanel';
import WikiTab from './memory/WikiTab';
import EvolutionTab from './memory/EvolutionTab';
import MemoryCenterTab from './memory/MemoryCenterTab';
import TimelineTab, { type TimelineTabProps } from './memory/TimelineTab';
import ImportReminderBanner from './memory/ImportReminderBanner';
@@ -23,7 +22,7 @@ import { useOnboarding } from '@/hooks/useOnboarding';
* Top-level structure is the ratified TWO-MIND SPLIT: "About you" reads the
* Personal Mind; "About this work" reads the active workspace's Mind. The
* per-mind list is MemoryCenterTab (reused, parameterized — D2). The legacy
* MemoryApp views (Timeline / Graph / Harvest / Weaver / Wiki / Evolution)
* MemoryApp views (Timeline / Graph / Harvest / Weaver / Wiki)
* survive as secondary tabs after a divider — capability preserved, entry
* restructured. Mind pills render only on the Memories tab: the legacy tabs
* keep their own scoping (Graph has a scope selector; Harvest writes to the
@@ -35,12 +34,12 @@ import { useOnboarding } from '@/hooks/useOnboarding';
*/
export type MindScope = 'personal' | 'workspace';
export type MemoryView = 'trust' | 'memories' | 'timeline' | 'graph' | 'harvest' | 'weaver' | 'wiki' | 'evolution';
export type MemoryView = 'trust' | 'memories' | 'timeline' | 'graph' | 'harvest' | 'weaver' | 'wiki';
// PR3.5: 'trust' (screen 19) is the new PRIMARY front door — first + default;
// the legacy views demote to secondary (after the divider).
export const MEMORY_VIEWS: readonly MemoryView[] = [
'trust', 'memories', 'timeline', 'graph', 'harvest', 'weaver', 'wiki', 'evolution',
'trust', 'memories', 'timeline', 'graph', 'harvest', 'weaver', 'wiki',
];
interface MemoryTabDef {
@@ -50,9 +49,9 @@ interface MemoryTabDef {
tooltip: string;
}
// UX gold-standard H1: 8 equal-weight tabs regrouped into 4 primary + an
// UX gold-standard H1: 7 equal-weight tabs regrouped into 4 primary + an
// "Advanced" flyout (display-level IA only — every view id, route and
// ?tab= deep-link is unchanged; MEMORY_VIEWS still validates all 8).
// ?tab= deep-link is unchanged; MEMORY_VIEWS still validates all 7).
const PRIMARY_TABS: MemoryTabDef[] = [
{ id: 'trust', label: 'Trust', icon: ShieldCheck, tooltip: 'Memory Trust — confidence & freshness, forget / correct / confirm, and "why did you do that?"' },
{ id: 'memories', label: 'Memories', icon: Brain, tooltip: 'Memory Center — inspect, edit, review, merge' },
@@ -64,7 +63,6 @@ const ADVANCED_TABS: MemoryTabDef[] = [
{ id: 'harvest', label: 'Imports', icon: Download, tooltip: 'Import conversations from other AIs' },
{ id: 'weaver', label: 'Maintenance', icon: Activity, tooltip: 'Memory distillation and consolidation' },
{ id: 'wiki', label: 'Wiki', icon: BookOpen, tooltip: 'Compiled knowledge pages' },
{ id: 'evolution', label: 'Improvements', icon: Sparkles, tooltip: 'Self-evolving prompts and agents' },
];
export interface MemoryCenterAppProps {
@@ -151,7 +149,7 @@ const MemoryCenterApp = ({
<div className="flex items-center">
<div className="w-px h-4 bg-border/60 mx-1" aria-hidden="true" />
<div className="relative">
<HintTooltip content="Imports, maintenance, wiki, and improvements">
<HintTooltip content="Imports, maintenance, and wiki">
<button
role="tab"
onClick={() => setAdvancedOpen(o => !o)}
@@ -279,10 +277,8 @@ const MemoryCenterApp = ({
<HarvestTab />
) : view === 'weaver' ? (
<WeaverPanel />
) : view === 'wiki' ? (
<WikiTab />
) : (
<EvolutionTab />
<WikiTab />
)}
</motion.div>
</div>

View File

@@ -1,253 +0,0 @@
/**
* EvolutionTab — PR6b B4 (skill-card + version-ladder view, D12/D13).
*
* Covers the new aggregation framing layered over the EXISTING run-review:
* - runs aggregate by (target_kind, target_name) into skill cards
* - the version ladder orders v1=oldest→vN=newest and renders score bars
* - the D13 provenance badge appears ONLY when a deployed run records a
* passing regression gate, and is OMITTED otherwise (no fabrication)
* - the "By skill" / "By run" toggle preserves the raw run list
* - selecting a version loads the existing detail view
*/
import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest';
import { render, screen, cleanup, fireEvent } from '@testing-library/react';
import { TooltipProvider } from '@/components/ui/tooltip';
const mocks = vi.hoisted(() => ({
adapter: {
fetch: vi.fn(),
},
}));
vi.mock('@/lib/adapter', () => ({ adapter: mocks.adapter, default: vi.fn() }));
import EvolutionTab from './EvolutionTab';
/** HintTooltip (radix) requires a provider in the tree. */
function renderTab() {
return render(
<TooltipProvider>
<EvolutionTab />
</TooltipProvider>,
);
}
interface FakeRun {
id: number;
run_uuid: string;
target_kind: string;
target_name: string | null;
baseline_text: string;
winner_text: string;
delta_accuracy: number;
gate_verdict: 'pass' | 'fail' | 'warn';
gate_reasons_json: string;
status: 'proposed' | 'accepted' | 'rejected' | 'deployed' | 'failed';
user_note?: string | null;
created_at: string;
}
const baseRun = (over: Partial<FakeRun>): FakeRun => ({
id: 1,
run_uuid: 'uuid-1',
target_kind: 'behavioral-spec-section',
target_name: 'Section A',
baseline_text: 'baseline',
winner_text: 'winner',
delta_accuracy: 0.05,
gate_verdict: 'pass',
gate_reasons_json: '[]',
status: 'proposed',
user_note: null,
created_at: '2026-06-01T00:00:00.000Z',
...over,
});
// Two versions of the same skill (Section A) + one different skill (Section B).
const RUNS: FakeRun[] = [
baseRun({
id: 1, run_uuid: 'a-v1', target_name: 'Section A',
delta_accuracy: 0.03, status: 'rejected', user_note: 'first draft',
created_at: '2026-06-01T00:00:00.000Z',
}),
baseRun({
id: 2, run_uuid: 'a-v2', target_name: 'Section A',
delta_accuracy: 0.09, status: 'deployed', user_note: 'added a citation lane',
gate_reasons_json: JSON.stringify([{ gate: 'regression', verdict: 'pass', reason: 'beat baseline' }]),
created_at: '2026-06-05T00:00:00.000Z',
}),
baseRun({
id: 3, run_uuid: 'b-v1', target_name: 'Section B',
delta_accuracy: 0.02, status: 'proposed',
created_at: '2026-06-03T00:00:00.000Z',
}),
];
const STATUS = {
counts: { proposed: 1, accepted: 0, rejected: 1, deployed: 1, failed: 0 },
pendingCount: 1,
};
function routeFetch(url: string): Response {
if (url.includes('/api/evolution/targets')) {
return {
ok: true,
json: async () => ({
personas: [{ id: 'general-purpose', name: 'General Purpose' }],
sections: ['Section A'],
defaultSchema: { kind: 'baseline' },
}),
} as unknown as Response;
}
if (url.includes('/api/evolution/baseline')) {
return {
ok: true,
json: async () => ({
baseline: 'section baseline text',
schemaBaseline: { kind: 'baseline' },
}),
} as unknown as Response;
}
if (url.includes('/api/evolution/status')) {
return { ok: true, json: async () => STATUS } as unknown as Response;
}
if (url.includes('/api/evolution/runs/')) {
const uuid = url.split('/api/evolution/runs/')[1];
const run = RUNS.find(r => r.run_uuid === uuid);
return {
ok: true,
json: async () => ({ ...run, winnerSchema: null, artifacts: null, gateReasons: [] }),
} as unknown as Response;
}
if (url.includes('/api/evolution/runs')) {
return { ok: true, json: async () => ({ runs: RUNS }) } as unknown as Response;
}
return { ok: true, json: async () => ({}) } as unknown as Response;
}
beforeEach(() => {
mocks.adapter.fetch.mockImplementation((url: string) =>
Promise.resolve(routeFetch(url)),
);
});
afterEach(() => { cleanup(); vi.clearAllMocks(); });
describe('EvolutionTab — skill-card / version-ladder (D12)', () => {
it('aggregates runs by target into skill cards (two cards for two distinct targets)', async () => {
renderTab();
// Skill view is the default — both targets appear as card headers.
expect(await screen.findByText('Section A')).toBeInTheDocument();
expect(screen.getByText('Section B')).toBeInTheDocument();
});
it('renders a version ladder ordered v1=oldest with score bars per version', async () => {
renderTab();
await screen.findByText('Section A');
// Subtitle text is interpolated ("<kind> · N versions") — match on the
// count substring with a node matcher.
expect(screen.getByText(/2 versions/)).toBeInTheDocument();
expect(screen.getByText(/1 version\b/)).toBeInTheDocument();
// v1/v2 tags + the human notes used as improve-notes, in order.
expect(screen.getByText('first draft')).toBeInTheDocument();
expect(screen.getByText('added a citation lane')).toBeInTheDocument();
expect(screen.getAllByText('v1').length).toBeGreaterThanOrEqual(2);
expect(screen.getByText('v2')).toBeInTheDocument();
});
it('shows the D13 provenance badge ONLY for the deployed run with a passing regression gate', async () => {
renderTab();
await screen.findByText('Section A');
// The deployed v2 (regression pass) earns the verified pill; nothing else does.
const badges = screen.getAllByText(/regression gate · score-verified/i);
expect(badges).toHaveLength(1);
});
it('does NOT fabricate a "judged by 3 models" / GEPA roster string', async () => {
renderTab();
await screen.findByText('Section A');
expect(screen.queryByText(/judged by 3 models/i)).not.toBeInTheDocument();
expect(screen.queryByText(/GEPA optimized/i)).not.toBeInTheDocument();
});
it('preserves the raw run list behind the "By run" toggle, with status filter chips', async () => {
renderTab();
await screen.findByText('Section A');
// Filter chips are hidden in skill view.
expect(screen.queryByRole('button', { name: /^all/i })).not.toBeInTheDocument();
fireEvent.click(screen.getByRole('tab', { name: /by run/i }));
// Now the filter chips (raw run list controls) are present.
expect(await screen.findByRole('button', { name: /^all/i })).toBeInTheDocument();
});
it('selecting a version routes to the existing run detail view', async () => {
renderTab();
const card = (await screen.findByText('Section A')).closest('div')!;
// Click the deployed v2 version row.
fireEvent.click(screen.getByText('added a citation lane'));
// Detail view loads the run (right pane shows the Baseline vs Winner diff).
expect(await screen.findByText(/baseline vs winner/i)).toBeInTheDocument();
expect(card).toBeTruthy();
});
it('a11y: proposal review note exposes stable form metadata and a visible focus style', async () => {
renderTab();
fireEvent.click(await screen.findByRole('button', { name: /v1 awaiting review/i }));
const note = await screen.findByLabelText(/note \(optional\)/i);
expect(note).toHaveAttribute('name', 'evolutionReviewNote');
expect(note).toHaveAttribute('autocomplete', 'off');
expect(note.className).toContain('focus-visible:ring-2');
});
it('still surfaces the pending-review banner and New Run affordance', async () => {
renderTab();
expect(await screen.findByText(/1 proposal awaiting review/i)).toBeInTheDocument();
expect(screen.getByRole('button', { name: /new evolution run/i })).toBeInTheDocument();
});
it('a11y: new run modal close icon has a name and visible focus style', async () => {
renderTab();
fireEvent.click(await screen.findByRole('button', { name: /new evolution run/i }));
const close = await screen.findByRole('button', { name: /close new evolution run/i });
expect(close.className).toContain('focus-visible:ring-2');
});
it('a11y: new run modal controls expose stable form metadata and visible focus styles', async () => {
renderTab();
fireEvent.click(await screen.findByRole('button', { name: /new evolution run/i }));
const kind = await screen.findByLabelText(/target kind/i);
expect(kind).toHaveAttribute('id', 'evolution-run-target-kind');
expect(kind).toHaveAttribute('name', 'evolutionTargetKind');
expect(kind).toHaveAttribute('autocomplete', 'off');
expect(kind.className).toContain('focus-visible:ring-2');
const name = screen.getByLabelText(/target name/i);
expect(name).toHaveAttribute('id', 'evolution-run-target-name');
expect(name).toHaveAttribute('name', 'evolutionTargetName');
expect(name).toHaveAttribute('autocomplete', 'off');
expect(name.className).toContain('focus-visible:ring-2');
const baseline = await screen.findByLabelText(/^baseline/i);
expect(baseline).toHaveAttribute('id', 'evolution-run-baseline');
expect(baseline).toHaveAttribute('name', 'evolutionBaseline');
expect(baseline).toHaveAttribute('autocomplete', 'off');
expect(baseline.className).toContain('focus-visible:ring-2');
fireEvent.click(screen.getByRole('button', { name: /advanced \(schema baseline\)/i }));
const schema = await screen.findByLabelText(/schema baseline json/i);
expect(schema).toHaveAttribute('id', 'evolution-run-schema');
expect(schema).toHaveAttribute('name', 'evolutionSchemaBaseline');
expect(schema).toHaveAttribute('autocomplete', 'off');
expect(schema.className).toContain('focus-visible:ring-2');
});
it('shows the empty state when there are no runs', async () => {
mocks.adapter.fetch.mockImplementation((url: string) => {
if (url.includes('/api/evolution/status')) {
return Promise.resolve({ ok: true, json: async () => ({ counts: {}, pendingCount: 0 }) } as unknown as Response);
}
return Promise.resolve({ ok: true, json: async () => ({ runs: [] }) } as unknown as Response);
});
renderTab();
expect(await screen.findByText(/your skills get better on their own/i)).toBeInTheDocument();
});
});

File diff suppressed because it is too large Load Diff

View File

@@ -108,7 +108,7 @@ function actionToResult(action: ResolvedAction): CommandResult {
* (idle) palette by naming the power-tool categories — no sidebar items added,
* no new persistence. Typing filters them away via the curated catalog above. */
const POWER_TOOL_HINTS: readonly string[] = [
'Automations', 'Files', 'Vault', 'Usage', 'Connectors', 'Evolution',
'Automations', 'Files', 'Vault', 'Usage', 'Connectors',
];
/* ── Permission prompt (C9 — reuses the chat ApprovalGate visual pattern) ──