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:
@@ -10,7 +10,7 @@ import Fastify, { type FastifyInstance, type FastifyRequest } from 'fastify';
|
||||
import cors from '@fastify/cors';
|
||||
import fastifyStatic from '@fastify/static';
|
||||
import websocket from '@fastify/websocket';
|
||||
import { MindDB, MultiMind, MultiMindCache, WorkspaceManager, WaggleConfig, createEmbeddingProvider, type EmbeddingProviderConfig, type EmbeddingProviderInstance, FrameStore, SessionStore, SuppressionStore, InstallAuditStore, CronStore, AwarenessLayer, VaultStore, SkillHashStore, OptimizationLogStore, ImprovementSignalStore, HarvestSourceStore, ClaudeCodeAdapter, reconcileIndexes, TeamSync, TelemetryStore, TELEMETRY_EVENTS, ExecutionTraceStore, EvolutionRunStore, harvestSetHash, type WorkspaceConfig } from '@waggle/core';
|
||||
import { MindDB, MultiMind, MultiMindCache, WorkspaceManager, WaggleConfig, createEmbeddingProvider, type EmbeddingProviderConfig, type EmbeddingProviderInstance, FrameStore, SessionStore, SuppressionStore, InstallAuditStore, CronStore, AwarenessLayer, VaultStore, SkillHashStore, HarvestSourceStore, ClaudeCodeAdapter, reconcileIndexes, TeamSync, TelemetryStore, TELEMETRY_EVENTS, ExecutionTraceStore, harvestSetHash, type WorkspaceConfig } from '@waggle/core';
|
||||
import { corsOriginAllowed } from './cors-config.js';
|
||||
import { getBoundTeamServer } from './team-server-binding.js';
|
||||
import { fetchTeamServer } from './team-server-egress.js';
|
||||
@@ -46,18 +46,14 @@ import {
|
||||
createBrowserTools,
|
||||
createLspTools,
|
||||
createCliTools,
|
||||
createInsightsTools,
|
||||
createConnectorSearchTools,
|
||||
createCrossWorkspaceTools,
|
||||
McpRuntime,
|
||||
McpToolRetriever,
|
||||
isWithinBudget,
|
||||
getRecentLogs,
|
||||
setPersonaDataDir,
|
||||
deliverCronResult,
|
||||
createDefaultDeliveryPreferences,
|
||||
buildActiveBehavioralSpec,
|
||||
loadBehavioralSpecOverrides,
|
||||
TraceRecorder,
|
||||
HarnessTraceBridge,
|
||||
safeFetch,
|
||||
@@ -108,7 +104,6 @@ import { homeRoutes } from './routes/home.js';
|
||||
import { onboardingRoutes } from './routes/onboarding.js';
|
||||
import { cronRoutes } from './routes/cron.js';
|
||||
import { notificationRoutes, emitNotification, emitSubagentStatus } from './routes/notifications.js';
|
||||
import { materialFingerprint } from './notification-gate.js';
|
||||
import { marketplaceDevRoutes } from './routes/marketplace-dev.js';
|
||||
import { marketplaceRoutes } from './routes/marketplace.js';
|
||||
import {
|
||||
@@ -129,7 +124,6 @@ import { personaRoutes } from './routes/personas.js';
|
||||
import { feedbackRoutes } from './routes/feedback.js';
|
||||
import { workflowRoutes } from './routes/workflows.js';
|
||||
import { workspaceTemplateRoutes } from './routes/workspace-templates.js';
|
||||
import { evolutionRoutes } from './routes/evolution.js';
|
||||
import { exportRoutes } from './routes/export.js';
|
||||
import { dataEraseRoutes } from './routes/data-erase.js';
|
||||
import { costRoutes } from './routes/cost.js';
|
||||
@@ -154,7 +148,6 @@ import { runChannelChatTurn } from './channels/chat-client.js';
|
||||
import { channelRoutes } from './channels/routes.js';
|
||||
import { isChannelPlatform } from './channels/types.js';
|
||||
import { WorkspaceTurnCoordinator } from './workspace-turn-coordinator.js';
|
||||
import { IdleSessionWatcher, readRecentTranscript, buildReviewInstruction, NOTHING_TO_DO } from './idle-watcher.js';
|
||||
import { DreamJournal } from './dream-journal.js';
|
||||
import { dreamRoutes } from './routes/dreams.js';
|
||||
import { oauthRoutes } from './routes/oauth.js';
|
||||
@@ -183,7 +176,6 @@ import {
|
||||
securityMiddleware,
|
||||
} from './security-middleware.js';
|
||||
import { LocalScheduler, makeRecordExecutionCallback } from './cron.js';
|
||||
import { EvolutionService, isEvolutionAutoEnabled } from './services/evolution-service.js';
|
||||
import {
|
||||
generateMorningBriefing,
|
||||
checkStaleWorkspaces,
|
||||
@@ -192,7 +184,6 @@ import {
|
||||
type ProactiveContext,
|
||||
type ProactiveMessage,
|
||||
} from './proactive-handlers.js';
|
||||
import { generateMonthlyAssessment, saveAssessmentToMind } from './monthly-assessment.js';
|
||||
import { MAX_WORKSPACE_SESSIONS, WorkspaceSessionManager } from './workspace-sessions.js';
|
||||
import { EventEmitter } from 'node:events';
|
||||
import { LocalJobStore } from './job-store.js';
|
||||
@@ -392,21 +383,12 @@ declare module 'fastify' {
|
||||
auditStore: import('@waggle/core').InstallAuditStore;
|
||||
cronStore: import('@waggle/core').CronStore;
|
||||
traceStore: import('@waggle/core').ExecutionTraceStore;
|
||||
evolutionStore: import('@waggle/core').EvolutionRunStore;
|
||||
/**
|
||||
* Active behavioral spec — baseline `BEHAVIORAL_SPEC` with any
|
||||
* accepted-and-deployed evolution overrides applied. Re-derived on
|
||||
* the 'behavioral-spec:reloaded' event.
|
||||
*/
|
||||
/** Active behavioral spec — the compiled `BEHAVIORAL_SPEC` baseline. */
|
||||
activeBehavioralSpec: ReturnType<typeof import('@waggle/agent').buildActiveBehavioralSpec>;
|
||||
vault: import('@waggle/core').VaultStore;
|
||||
embeddingProvider: import('@waggle/core').EmbeddingProviderInstance;
|
||||
skillHashStore: import('@waggle/core').SkillHashStore;
|
||||
scheduler: import('./cron.js').LocalScheduler;
|
||||
/** Idle-session self-evolution watcher. Inert unless self-evolution.json enabled. */
|
||||
idleWatcher: import('./idle-watcher.js').IdleSessionWatcher;
|
||||
/** H-10 G1: autonomous evolution daemon. Inert unless WAGGLE_EVOLUTION_AUTO_ENABLED=1. */
|
||||
evolutionService: import('./services/evolution-service.js').EvolutionService;
|
||||
marketplace: import('@waggle/marketplace').MarketplaceDB | null;
|
||||
/** Shared per-workspace mind cache. Routes use this to reach an arbitrary
|
||||
* workspace's mind (e.g. FileIndexer for L-20 auto-indexing). */
|
||||
@@ -418,7 +400,7 @@ declare module 'fastify' {
|
||||
sessionManager: import('./workspace-sessions.js').WorkspaceSessionManager;
|
||||
/**
|
||||
* Bridges workflow-harness phase events into the execution trace store
|
||||
* so harnessed runs become training signal for the self-evolution loop.
|
||||
* so harnessed runs show up in cost tracking and run history.
|
||||
* Started at boot, stopped on close.
|
||||
*/
|
||||
harnessTraceBridge: HarnessTraceBridge;
|
||||
@@ -533,14 +515,14 @@ export async function buildLocalServer(config: Partial<LocalConfig> = {}) {
|
||||
|
||||
seedDefaultCrons(cronStore);
|
||||
|
||||
// Execution trace store — persistent log of agent runs (self-evolution substrate)
|
||||
// Execution trace store — persistent log of agent runs (cost tracking, run history)
|
||||
const traceStore = new ExecutionTraceStore(multiMind.personal);
|
||||
server.decorate('traceStore', traceStore);
|
||||
|
||||
// Workflow-harness → trace bridge. Any harness phase that completes or
|
||||
// aborts anywhere in the app emits on the shared `harnessEvents` emitter;
|
||||
// this bridge translates those into rows in `execution_traces` so harness
|
||||
// runs become training data for the self-evolution loop.
|
||||
// runs show up in cost tracking.
|
||||
const harnessTraceBridge = new HarnessTraceBridge({
|
||||
recorder: new TraceRecorder(traceStore),
|
||||
});
|
||||
@@ -550,22 +532,7 @@ export async function buildLocalServer(config: Partial<LocalConfig> = {}) {
|
||||
harnessTraceBridge.stop();
|
||||
});
|
||||
|
||||
// Evolution run store — audit trail of proposed/accepted/rejected self-evolution runs
|
||||
const evolutionStore = new EvolutionRunStore(multiMind.personal);
|
||||
server.decorate('evolutionStore', evolutionStore);
|
||||
|
||||
// Behavioral spec with deployed overrides applied (self-evolution deploy target).
|
||||
// Re-derived on the 'behavioral-spec:reloaded' event emitted by evolution routes.
|
||||
const initialOverrides = loadBehavioralSpecOverrides(fullConfig.dataDir);
|
||||
let activeBehavioralSpec = buildActiveBehavioralSpec(initialOverrides);
|
||||
server.decorate('activeBehavioralSpec', activeBehavioralSpec);
|
||||
eventBus.on('behavioral-spec:reloaded', () => {
|
||||
const latest = loadBehavioralSpecOverrides(fullConfig.dataDir);
|
||||
activeBehavioralSpec = buildActiveBehavioralSpec(latest);
|
||||
// Re-decorate so server.activeBehavioralSpec returns the fresh value.
|
||||
// Fastify's decorate() is idempotent for reassignment via the instance.
|
||||
(server as unknown as { activeBehavioralSpec: typeof activeBehavioralSpec }).activeBehavioralSpec = activeBehavioralSpec;
|
||||
});
|
||||
server.decorate('activeBehavioralSpec', buildActiveBehavioralSpec());
|
||||
|
||||
// Vault — encrypted secret storage
|
||||
const vault = new VaultStore(fullConfig.dataDir);
|
||||
@@ -819,86 +786,13 @@ export async function buildLocalServer(config: Partial<LocalConfig> = {}) {
|
||||
// Dynamic connector tools (initial — regenerated per workspace in buildToolsForWorkspace)
|
||||
const defaultConnectorTools = connectorRegistry.generateTools();
|
||||
|
||||
// Insights tools — self-analytics so the agent can reflect on its own performance.
|
||||
// Aggregates optimization logs + improvement signals across the personal mind and
|
||||
// every workspace so the agent sees a complete picture regardless of where it runs.
|
||||
const insightsTools = createInsightsTools({
|
||||
getOptimizationLogs: (limit: number) => {
|
||||
const rows: Array<{
|
||||
tools_used: string;
|
||||
was_correction: number;
|
||||
input_tokens: number;
|
||||
output_tokens: number;
|
||||
timestamp: string;
|
||||
}> = [];
|
||||
try {
|
||||
rows.push(...new OptimizationLogStore(multiMind.personal).getRecent(limit));
|
||||
} catch { /* personal store unavailable — keep going */ }
|
||||
for (const ws of wsManager.list()) {
|
||||
const wsDb = getWorkspaceMindDb(ws.id);
|
||||
if (!wsDb) continue;
|
||||
try {
|
||||
rows.push(...new OptimizationLogStore(wsDb).getRecent(limit));
|
||||
} catch { /* workspace store unavailable — skip */ }
|
||||
}
|
||||
rows.sort((a, b) => (b.timestamp ?? '').localeCompare(a.timestamp ?? ''));
|
||||
return rows.slice(0, limit).map(row => {
|
||||
let toolNames = '';
|
||||
try {
|
||||
const parsed = JSON.parse(row.tools_used ?? '[]');
|
||||
if (Array.isArray(parsed)) toolNames = parsed.join(',');
|
||||
} catch { /* malformed — leave empty */ }
|
||||
return {
|
||||
model: undefined,
|
||||
tool_names: toolNames,
|
||||
total_tokens: (row.input_tokens ?? 0) + (row.output_tokens ?? 0),
|
||||
estimated_cost: 0,
|
||||
was_correction: row.was_correction,
|
||||
created_at: row.timestamp,
|
||||
};
|
||||
});
|
||||
},
|
||||
getImprovementSignals: () => {
|
||||
const collectFrom = (db: MindDB) => {
|
||||
const store = new ImprovementSignalStore(db);
|
||||
return [
|
||||
...store.getByCategory('capability_gap'),
|
||||
...store.getByCategory('correction'),
|
||||
...store.getByCategory('workflow_pattern'),
|
||||
];
|
||||
};
|
||||
const all: Array<{
|
||||
category: string;
|
||||
pattern_key: string;
|
||||
detail: string;
|
||||
count: number;
|
||||
first_seen: string;
|
||||
last_seen: string;
|
||||
}> = [];
|
||||
try { all.push(...collectFrom(multiMind.personal)); } catch { /* skip */ }
|
||||
for (const ws of wsManager.list()) {
|
||||
const wsDb = getWorkspaceMindDb(ws.id);
|
||||
if (!wsDb) continue;
|
||||
try { all.push(...collectFrom(wsDb)); } catch { /* skip */ }
|
||||
}
|
||||
return all.map(s => ({
|
||||
category: s.category,
|
||||
pattern_key: s.pattern_key,
|
||||
detail: s.detail,
|
||||
count: s.count,
|
||||
first_seen: s.first_seen,
|
||||
last_seen: s.last_seen,
|
||||
}));
|
||||
},
|
||||
});
|
||||
|
||||
// Connector search tools — give the agent natural-language discovery over
|
||||
// the 148-entry MCP catalog (find_connector, list_connector_categories).
|
||||
// Static data, zero I/O, no deps — cheap to always-include.
|
||||
const connectorSearchTools = createConnectorSearchTools();
|
||||
|
||||
// Collect all non-subagent tools first (sub-agent tools need the full list)
|
||||
const baseTools = [...mindTools, ...systemTools, ...planTools, ...gitTools, ...documentTools, ...skillTools, ...cronTools, ...searchTools, ...browserTools, ...lspTools, ...cliTools, ...insightsTools, ...connectorSearchTools, ...defaultConnectorTools];
|
||||
const baseTools = [...mindTools, ...systemTools, ...planTools, ...gitTools, ...documentTools, ...skillTools, ...cronTools, ...searchTools, ...browserTools, ...lspTools, ...cliTools, ...connectorSearchTools, ...defaultConnectorTools];
|
||||
|
||||
// PR4: expose the full native-tool name union to the agent-search route so
|
||||
// its "you already have a tool for this" lane isn't blind to search/browser/
|
||||
@@ -945,7 +839,7 @@ export async function buildLocalServer(config: Partial<LocalConfig> = {}) {
|
||||
const targetDb = wsDb ?? multiMind.personal;
|
||||
const mindLabel = wsDb ? `workspace:${wsId}` : 'personal';
|
||||
|
||||
// Structured content: tag provenance so search/evolution can filter
|
||||
// Structured content: tag provenance so search can filter
|
||||
// for sub-agent-derived frames and so humans can see where it came from.
|
||||
const content = `[Sub-agent: ${result.role}] ${result.agentName}\n\n${result.response}\n\n— tools: ${result.toolsUsed.join(', ') || 'none'} · ${(result.duration / 1000).toFixed(1)}s · ${result.usage.inputTokens + result.usage.outputTokens} tokens`;
|
||||
|
||||
@@ -1874,7 +1768,7 @@ export async function buildLocalServer(config: Partial<LocalConfig> = {}) {
|
||||
// W4.3d: extraction of the benchmark-proven recall lanes
|
||||
// (facts/events/profiles) over frames written since the last run.
|
||||
// Routes through the built-in proxy ('fast' tier = haiku) — the
|
||||
// same internal-LLM pattern as prompt_optimization above.
|
||||
// same internal-LLM pattern as the other cron LLM calls.
|
||||
try {
|
||||
const llmCall = async (prompt: string, _model: 'fast' | 'accurate'): Promise<string> => {
|
||||
const proxyUrl = `http://127.0.0.1:${fullConfig.port ?? 3333}/v1/chat/completions`;
|
||||
@@ -2041,143 +1935,6 @@ export async function buildLocalServer(config: Partial<LocalConfig> = {}) {
|
||||
}
|
||||
break;
|
||||
}
|
||||
case 'prompt_optimization': {
|
||||
// Background GEPA/Ax prompt optimization — runs daily at 2 AM.
|
||||
// Iterates workspaces with optimizationEnabled, checks budget, reads logs,
|
||||
// and invokes PromptOptimizer for variant generation.
|
||||
try {
|
||||
const workspaces = wsManager.list().filter(ws => ws.optimizationEnabled);
|
||||
for (const ws of workspaces) {
|
||||
const wsDb = getWorkspaceMindDb(ws.id);
|
||||
if (!wsDb) continue;
|
||||
|
||||
const optStore = new OptimizationLogStore(wsDb);
|
||||
const budget = ws.optimizationBudget ?? 100; // cents, default $1/day
|
||||
|
||||
// Check budget before spending any tokens
|
||||
if (!isWithinBudget(optStore, budget)) {
|
||||
log.info(`[cron] Prompt optimization: workspace "${ws.name}" over budget, skipping`);
|
||||
continue;
|
||||
}
|
||||
|
||||
// Read recent logs for analysis
|
||||
const recentLogs = getRecentLogs(optStore, 100);
|
||||
if (recentLogs.length < 5) {
|
||||
// Not enough data to optimize yet
|
||||
continue;
|
||||
}
|
||||
|
||||
// Analyze patterns: high correction rate, common tools, avg turn count
|
||||
const stats = optStore.getStats();
|
||||
const correctionRate = stats.correctionRate;
|
||||
const avgTurns = stats.avgTurnCount;
|
||||
|
||||
// Only trigger optimization if correction rate is above threshold
|
||||
// or turn count is significantly above average (indicating inefficiency)
|
||||
if (correctionRate > 0.2 || avgTurns > 15) {
|
||||
const rate = (correctionRate * 100).toFixed(1);
|
||||
log.info(`[cron] Prompt optimization: workspace "${ws.name}" — correction rate ${rate}%, avg turns ${avgTurns.toFixed(1)}`);
|
||||
|
||||
// Re-check budget before the LLM call (another workspace may have consumed tokens)
|
||||
if (!isWithinBudget(optStore, budget)) {
|
||||
log.info(` GEPA optimization skipped — daily budget exceeded`);
|
||||
continue;
|
||||
}
|
||||
|
||||
// Get the most recent system prompt from the workspace's logs
|
||||
const currentSystemPrompt = recentLogs[0]?.system_prompt ?? '';
|
||||
if (!currentSystemPrompt) {
|
||||
log.info(`[cron] Prompt optimization: no system prompt in logs, skipping`);
|
||||
continue;
|
||||
}
|
||||
|
||||
// Generate variant via the built-in Anthropic proxy (localhost)
|
||||
try {
|
||||
const proxyUrl = `http://127.0.0.1:${fullConfig.port ?? 3333}/v1/chat/completions`;
|
||||
const variantResponse = await fetch(proxyUrl, {
|
||||
method: 'POST',
|
||||
headers: { 'Content-Type': 'application/json' },
|
||||
body: JSON.stringify({
|
||||
model: 'claude-sonnet-4-6',
|
||||
max_tokens: 4096,
|
||||
messages: [
|
||||
{
|
||||
role: 'system',
|
||||
content: `You are a prompt optimization specialist. Your task is to improve an AI agent's system prompt based on performance signals.
|
||||
|
||||
The agent's correction rate is ${rate}% (target: under 20%) and average turn count is ${avgTurns.toFixed(1)} (target: under 15).
|
||||
|
||||
A high correction rate means the agent frequently misunderstands instructions or produces wrong outputs.
|
||||
A high turn count means the agent takes too many steps to complete tasks.
|
||||
|
||||
Analyze the current system prompt and generate an improved variant that:
|
||||
1. Adds clearer instructions for areas where corrections are common
|
||||
2. Reduces unnecessary verbosity that inflates turn count
|
||||
3. Preserves the core identity and capabilities
|
||||
4. Is production-ready (not a draft or explanation)
|
||||
|
||||
Return ONLY the improved system prompt text. No commentary, no markdown fences, no explanation.`,
|
||||
},
|
||||
{
|
||||
role: 'user',
|
||||
content: `Here is the current system prompt to improve:\n\n${currentSystemPrompt.slice(0, 12000)}`,
|
||||
},
|
||||
],
|
||||
}),
|
||||
signal: AbortSignal.timeout(60_000),
|
||||
});
|
||||
|
||||
if (!variantResponse.ok) {
|
||||
log.warn(`[cron] GEPA variant generation failed: HTTP ${variantResponse.status}`);
|
||||
} else {
|
||||
const variantBody = await variantResponse.json() as unknown;
|
||||
const variantText = parseOpenAiTextCompletion(variantBody).content;
|
||||
|
||||
if (variantText.length > 100) {
|
||||
// Store the variant in the optimization_log with a marker
|
||||
optStore.insert({
|
||||
sessionId: `gepa-variant-${Date.now()}`,
|
||||
workspaceId: ws.id,
|
||||
systemPrompt: variantText,
|
||||
toolsUsed: ['gepa_variant'],
|
||||
turnCount: 0,
|
||||
wasCorrection: false,
|
||||
inputTokens: currentSystemPrompt.length,
|
||||
outputTokens: variantText.length,
|
||||
});
|
||||
log.info(` GEPA generated prompt variant (correction_rate=${rate}%)`);
|
||||
} else {
|
||||
log.warn(`[cron] GEPA variant too short (${variantText.length} chars), discarding`);
|
||||
}
|
||||
}
|
||||
} catch (variantErr) {
|
||||
log.warn(`[cron] GEPA variant generation error: ${(variantErr as Error).message}`);
|
||||
}
|
||||
|
||||
// Anti-nag: one signal per (workspace, month, correction-rate bucket).
|
||||
// Re-runs on the same month with an unchanged rate bucket are suppressed.
|
||||
emitNotification(server, {
|
||||
title: 'Prompt optimization signal',
|
||||
body: `Workspace "${ws.name}" has ${(correctionRate * 100).toFixed(0)}% correction rate — optimization candidate`,
|
||||
category: 'agent',
|
||||
}, {
|
||||
dedupeKey: `prompt_optimization:${ws.id}`,
|
||||
materialHash: materialFingerprint({
|
||||
wsId: ws.id,
|
||||
period: new Date().toISOString().slice(0, 7),
|
||||
rateBucket: Math.round(correctionRate * 10),
|
||||
}),
|
||||
});
|
||||
}
|
||||
|
||||
// Prune old logs (>30 days) to keep the .mind file lean
|
||||
optStore.pruneOlderThan(30);
|
||||
}
|
||||
} catch (err) {
|
||||
log.warn(`[cron] Prompt optimization failed: ${(err as Error).message}`);
|
||||
}
|
||||
break;
|
||||
}
|
||||
case 'agent_task': {
|
||||
// F9: Execute an agent task prompt in a workspace context
|
||||
const taskConfig = JSON.parse(schedule.job_config || '{}');
|
||||
@@ -2363,31 +2120,6 @@ Return ONLY the improved system prompt text. No commentary, no markdown fences,
|
||||
}
|
||||
break;
|
||||
}
|
||||
case 'monthly_assessment': {
|
||||
try {
|
||||
const assessment = generateMonthlyAssessment(fullConfig, multiMind.personal);
|
||||
saveAssessmentToMind(multiMind.personal, assessment);
|
||||
log.info(`[cron] Monthly assessment for ${assessment.period}: ${assessment.totalInteractions} interactions, ${(assessment.correctionRate * 100).toFixed(1)}% correction rate`);
|
||||
emitNotification(server, {
|
||||
title: 'Monthly agent assessment ready',
|
||||
body: `${assessment.period} report: ${assessment.totalInteractions} interactions, ${(assessment.correctionRate * 100).toFixed(1)}% correction rate — check your workspace home`,
|
||||
category: 'agent',
|
||||
actionUrl: '/',
|
||||
}, {
|
||||
// Anti-nag: one per period + counters. Same month with unchanged
|
||||
// counters is suppressed; a new month or changed numbers re-fires.
|
||||
dedupeKey: 'monthly_assessment',
|
||||
materialHash: materialFingerprint({
|
||||
period: assessment.period,
|
||||
totalInteractions: assessment.totalInteractions,
|
||||
correctionRateBucket: Math.round(assessment.correctionRate * 100),
|
||||
}),
|
||||
});
|
||||
} catch (err) {
|
||||
log.warn(`[cron] Monthly assessment failed: ${(err as Error).message}`);
|
||||
}
|
||||
break;
|
||||
}
|
||||
case 'loop': {
|
||||
// Loop v0 (L1, report-only): a stateful, memory-powered scheduled
|
||||
// automation. Resolve the target mind (a specific workspace, else the
|
||||
@@ -2509,66 +2241,6 @@ Return ONLY the improved system prompt text. No commentary, no markdown fences,
|
||||
scheduler.start();
|
||||
server.decorate('scheduler', scheduler);
|
||||
|
||||
// Self-evolution: idle-session watcher (CowAgent steal #3, v1). Default OFF via
|
||||
// <dataDir>/self-evolution.json; when enabled, it reviews idle sessions through
|
||||
// a restricted `session-reviewer` loopback turn whose only write (create_skill)
|
||||
// is HELD for human approval. Skipped under test to keep unit runs hermetic
|
||||
// (mirrors the ChannelManager VITEST guard below).
|
||||
const idleWatcher = new IdleSessionWatcher({
|
||||
dataDir: fullConfig.dataDir,
|
||||
emitNotification: (event, options) => emitNotification(server, event, options),
|
||||
log: { info: (msg: string) => log.info(msg), warn: (msg: string) => log.warn(msg) },
|
||||
runReviewTurn: async ({ sessionId, workspaceId }) => {
|
||||
const transcript = readRecentTranscript(fullConfig.dataDir, workspaceId, sessionId);
|
||||
if (!transcript) return { content: NOTHING_TO_DO };
|
||||
const res = await runChannelChatTurn({
|
||||
port: server.localConfig.port,
|
||||
sessionToken: server.agentState.wsSessionToken,
|
||||
message: buildReviewInstruction(sessionId, transcript),
|
||||
workspace: workspaceId,
|
||||
session: `evolve-${sessionId}`,
|
||||
persona: 'session-reviewer',
|
||||
proposeHeld: true,
|
||||
origin: 'automation',
|
||||
});
|
||||
return { content: res.content, ...(res.error ? { error: res.error } : {}) };
|
||||
},
|
||||
});
|
||||
server.decorate('idleWatcher', idleWatcher);
|
||||
if (!process.env.VITEST && process.env.NODE_ENV !== 'test') {
|
||||
idleWatcher.start();
|
||||
}
|
||||
server.addHook('onClose', async () => {
|
||||
idleWatcher.stop();
|
||||
});
|
||||
|
||||
// H-10 G1: Evolution service — autonomous self-evolution daemon.
|
||||
// Disabled by default; opt-in via WAGGLE_EVOLUTION_AUTO_ENABLED=1 so
|
||||
// existing deployments don't start burning Anthropic credits silently.
|
||||
// Even when enabled, the service only produces `proposed` runs — accept
|
||||
// still happens manually via the Memory → Evolution UI.
|
||||
const evolutionService = new EvolutionService(
|
||||
{
|
||||
traceStore,
|
||||
runStore: evolutionStore,
|
||||
getApiKey: () => server.vault?.get('anthropic')?.value ?? null,
|
||||
getActiveBehavioralSpec: () => (server as unknown as { activeBehavioralSpec: Record<string, unknown> }).activeBehavioralSpec,
|
||||
log: (level, msg) => {
|
||||
if (level === 'warn') log.warn(`[evolution-service] ${msg}`);
|
||||
else log.info(`[evolution-service] ${msg}`);
|
||||
},
|
||||
},
|
||||
{
|
||||
tickIntervalMs: parseInt(process.env.WAGGLE_EVOLUTION_TICK_INTERVAL_MS ?? '', 10) || undefined,
|
||||
minTracesPerTarget: parseInt(process.env.WAGGLE_EVOLUTION_MIN_TRACES ?? '', 10) || undefined,
|
||||
},
|
||||
);
|
||||
server.decorate('evolutionService', evolutionService);
|
||||
if (isEvolutionAutoEnabled()) {
|
||||
evolutionService.start();
|
||||
log.info('[evolution-service] autonomous evolution enabled (WAGGLE_EVOLUTION_AUTO_ENABLED=1)');
|
||||
}
|
||||
|
||||
// F2: Daily audit event cleanup (retain 90 days by default)
|
||||
const auditRetentionDays = parseInt(process.env.WAGGLE_AUDIT_RETENTION_DAYS ?? '90', 10);
|
||||
const auditCleanupTimer = setInterval(() => {
|
||||
@@ -2807,7 +2479,6 @@ Return ONLY the improved system prompt text. No commentary, no markdown fences,
|
||||
await server.register(personaRoutes);
|
||||
await server.register(feedbackRoutes);
|
||||
await server.register(workspaceTemplateRoutes);
|
||||
await server.register(evolutionRoutes);
|
||||
await server.register(exportRoutes);
|
||||
await server.register(dataEraseRoutes);
|
||||
await server.register(costRoutes);
|
||||
@@ -3288,7 +2959,6 @@ Return ONLY the improved system prompt text. No commentary, no markdown fences,
|
||||
agentRunRegistry.close();
|
||||
stopMarketplaceBackgroundSync();
|
||||
scheduler.stop();
|
||||
evolutionService.stop();
|
||||
await offlineManager.stop();
|
||||
clearInterval(auditCleanupTimer);
|
||||
|
||||
|
||||
Reference in New Issue
Block a user