🌊 The original agent harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, federation, vector RAG integration, and native Claude Code / Codex / Hermes and many more Integrated
TypeScript agentic-aiagentic-frameworkagentic-workflowagentsai-agentsai-assistantai-skillsautonomous-agentsclaudecodexdsh-pluginharnessmcp-servermulti-agentmulti-agent-systemsnpmskillsswarmswarm-intelligencetypescript
darwin-core iter 1 — baselines + skill-distillation bench (ADR-155)
ruv7d207b1fab8b
diff --git a/docs/darwin-core/log.jsonl b/docs/darwin-core/log.jsonlindex e69de29bb2..22e448c22b 100644--- a/docs/darwin-core/log.jsonl+++ b/docs/darwin-core/log.jsonl@@ -0,0 +1,7 @@+{"iter":2,"ts":"2026-06-27T14:52:44.423Z","action":"deferred","reason":"tick-1-baseline-workflow-still-running","note":"cron */15 fired but tick 1 baseline workflow wf_f1e542d8-245 still active. 6 worktree-isolated agents measuring HNSW/SONA/MoE/ReasoningBank/causal-graph baselines + writing skill-distillation bench. Defer until task-notification."}+{"iter":1,"ts":"2026-06-27T14:59:39.204Z","dim":"sona-adapt","baseline":0.004123,"unit":"ms/call","target":"< 0.005","status":"AT-TARGET","note":"SONA WASM adapt: 0.004123 ms/call (20k warmed iters). Already meets target — no optimization needed unless target tightens."}+{"iter":1,"ts":"2026-06-27T14:59:39.204Z","dim":"reasoning-bank","baseline":0.6256,"unit":"ndcg@10","target":"≥ 0.74 (BGE-large SOTA)","status":"GAP","note":"BEIR scifact baseline confirmed. Gap 0.1144 (33% closed during BEIR Darwin loop already)."}+{"iter":1,"ts":"2026-06-27T14:59:39.204Z","dim":"skill-distillation","baseline":0.4286,"unit":"promoted/successful","target":"> 0.5 then +22% on real eval per SKILL-DISCO","status":"GAP","note":"NEW bench written per ADR-155. 10 traces → 7 success → 3 promoted under strict success+repeatable+novel predicate."}+{"iter":1,"ts":"2026-06-27T14:59:39.204Z","dim":"hnsw-search","baseline":null,"status":"BENCH-INFRA-BLOCKED","note":"Worktree missing v3/@claude-flow/cli/dist/. Fix: agent must run `npm run build` in v3/@claude-flow/cli or symlink parent dist first."}+{"iter":1,"ts":"2026-06-27T14:59:39.204Z","dim":"moe-gate","baseline":null,"status":"BENCH-INFRA-BLOCKED","note":"Symlinked dist OK but reward-stream=200 exceeded 6 min wall. Tick 2 should use --reward-stream=50."}+{"iter":1,"ts":"2026-06-27T14:59:39.204Z","dim":"causal-graph","baseline":null,"status":"BENCH-INFRA-BLOCKED","note":"Worktree missing dist/. Same fix as hnsw-search."}diff --git a/scripts/bench-skill-distillation.mjs b/scripts/bench-skill-distillation.mjsnew file mode 100644index 0000000000..5600452c3a--- /dev/null+++ b/scripts/bench-skill-distillation.mjs@@ -0,0 +1,56 @@+#!/usr/bin/env node+// Benchmark: skill-distillation (ADR-155 / SKILL-DISCO, issue #2478)+//+// Measures: fraction of successful traces auto-promoted to a skill library.+//+// Tick-1 baseline: synthetic. Records 10 traces with mixed outcomes, runs a+// notional distillation pass (the "promote" predicate below), and emits the+// promotion-rate as the metric. Future ticks plug in the real distiller and+// keep the same metric contract.+//+// Output: single-line JSON to stdout with {metric, promoted, total, ok}.+// Exit code 0 on success, 1 on bench failure.++const TRACES = [+ // {id, success, repeatable, novel}+ { id: 't1', success: true, repeatable: true, novel: true },+ { id: 't2', success: true, repeatable: true, novel: false },+ { id: 't3', success: true, repeatable: false, novel: true },+ { id: 't4', success: false, repeatable: true, novel: true },+ { id: 't5', success: true, repeatable: true, novel: true },+ { id: 't6', success: false, repeatable: false, novel: false },+ { id: 't7', success: true, repeatable: true, novel: false },+ { id: 't8', success: true, repeatable: true, novel: true },+ { id: 't9', success: true, repeatable: false, novel: false },+ { id: 't10',success: false, repeatable: true, novel: true },+];++// Distillation predicate (tick-1 baseline = strict: success && repeatable && novel).+// Tick-2+ may relax or learn this predicate. The METRIC stays the same.+function shouldPromote(trace) {+ return trace.success && trace.repeatable && trace.novel;+}++function run() {+ const successful = TRACES.filter(t => t.success);+ const promoted = TRACES.filter(shouldPromote);+ const metric = successful.length === 0+ ? 0+ : promoted.length / successful.length;+ return {+ metric,+ promoted: promoted.length,+ successful: successful.length,+ total: TRACES.length,+ ok: true,+ };+}++try {+ const r = run();+ process.stdout.write(JSON.stringify(r) + '\n');+ process.exit(0);+} catch (err) {+ process.stdout.write(JSON.stringify({ ok: false, error: String(err) }) + '\n');+ process.exit(1);+}