본문으로 건너뛰기
cccskills
무료GitHub에 공개

evolve

Improve an AI agent's harness, the software that organizes its work, while keeping the model fixed. Test changes in a sandbox, measure their results, and retain variants that pass tests and demonstrate improvement.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

원문원문 설명 보기

Evolve this harness with Darwin Mode — frozen model, evolving harness (real, sandboxed, safety-gated).

ruvnet/RuView
설치 방법 보기

이럴 때 유용합니다

  • Tune an agent harness without changing its model
  • Compare changes to planning, context, or retry policies
  • Try an offline dry run without executing tests
  • Archive variants that pass tests and improve measured results

포함된 파일(1)

  • SKILL.md2.7 KB

SKILL.md(원문)

설치하기 전에 에이전트에게 전달되는 지시 내용을 확인할 수 있습니다.

evolve — Darwin Mode self-improvement

wifi-densepose-sar-harness ships with Darwin Mode (@metaharness/darwin, ADR-070…146): the model is frozen; the harness evolves. Each generation mutates ONE of the 7 surface files (planner, contextBuilder, reviewer, retry/tool/memory/score policy), sandboxes each child, scores it, and keeps only variants that measurably improve — building an archive of successful descendants.

Run it

npm run evolve        # real substrate: runs your test command per variant (deterministic mutator — no API key, no network)
npm run evolve:dry    # mock substrate: fast, fully offline, no test execution

Or directly:

npx metaharness-darwin evolve . --sandbox real --generations 3 --children 4

Safety (secure by default)

  • Deterministic mutator is the default — no network, no API key, air-gapped.
  • Every mutation passes the validateGeneratedCode gate: no new imports, network, filesystem, shell, env access, or dependencies — pure refactor/tuning only.
  • Mutations run in a sandbox; only variants that pass your tests are archived.
  • Nothing is promoted without measured improvement (guard against Goodharting).

See @metaharness/darwin for selection strategies (--selection, --crossover, --curriculum), statistical gates (--fdr, --bench), and the real-LLM mutator (library API).

What the benchmarks taught us (measured, full SWE-bench Lite 300)

Defaults worth carrying into how you evolve and run this harness (full evidence + CIs in @metaharness/darwin's LEARNINGS.md / bench/results/RESULTS.md):

  1. Closed-loop repair is the #1 lever (~2×). Feeding test/compiler failure back and retrying took resolve-rate 7.7% → 15.3% on the same cheap model. Iterate against ground truth, don't single-shot.
  2. Cheap-first + cost-aware routing. Track $/resolve, not just resolve-rate; a cheap model resolved 31× cheaper per fix than a frontier one. Reserve frontier for measured capability gaps.
  3. Tier the models (Barbarian & Scholar). Cheap sweep + frontier on only the residual = 33.3% at ~6× lower cost than running frontier everywhere.
  4. Put the output-format contract in a system message + example, and size prompts to the model's real context window — this alone took a weak local model from 0% to ~50% valid output.
  5. Only trust batch evaluation of the final artifact — in-loop counters drift 1.5–5×.
  6. The harness multiplies the model; it can't rescue one below the task's reasoning floor. Pick the smallest model above the floor, then let evolution do the rest.

리뷰

아직 리뷰가 없습니다. 사용해 본 소감을 남겨 주세요.

댓글

로그인 없이 쓸 수 있습니다. 사용 후기, 질문, 작동하지 않은 환경 등을 편하게 남겨 주세요.

아직 댓글이 없습니다. 첫 댓글을 남겨 보세요.

보내면 바로 공개됩니다. 이메일 주소나 전화번호 같은 개인 정보는 쓰지 마세요. URL은 2개까지입니다.

0 / 1,000자

이 양식은 reCAPTCHA로 보호되며 Google 개인정보처리방침 및 서비스 약관이 적용됩니다.

같은 저장소의 스킬

AgentDB Advanced Features

무료

Build advanced AgentDB workflows with database synchronization, multiple databases, and search that combines meaning with attribute filters. Covers similarity measures, diverse search results, and operational patterns for distributed AI applications.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

  • Sharing stored patterns across multiple AgentDB instances
  • Filtering semantic search by date or category
  • Separating knowledge and conversation databases
ruvnet/RuView9.9만2026년 10월 19일 업데이트

AgentDB Learning Plugins

무료

Create AgentDB learning plugins that use recorded experience to improve agent decisions. Choose a learning method, store states, actions, and rewards, train models, and evaluate suggested actions through CLI and API workflows.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

  • Train agents from historical experience
  • Build decision-making for games and navigation
  • Combine experience from multiple agents for training
ruvnet/RuView9.9만2026년 10월 19일 업데이트

AgentDB Memory Patterns

무료

Build AI agents that retain conversation history, user preferences, and useful interaction patterns with AgentDB. The skill covers persistent storage, meaning memories survive between sessions, plus retrieval, context management, and ReasoningBank integration.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

  • Add conversation history to a chat system
  • Remember user preferences across sessions
  • Reuse patterns from successful interactions
ruvnet/RuView9.9만2026년 10월 19일 업데이트

AgentDB Performance Optimization

무료

Tune AgentDB memory use and search speed through quantization, which stores vectors more compactly, search indexes, caching, and batch operations. Includes configuration examples and benchmarks to help balance accuracy, speed, and storage as datasets grow.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

  • Reduce memory use in an AgentDB application
  • Tune search speed while preserving accuracy
  • Plan settings for millions of vectors
ruvnet/RuView9.9만2026년 10월 19일 업데이트

AgentDB Vector Search

무료

Build document search with AgentDB that finds related content by meaning. Store documents as numerical representations, combine similarity search with metadata filters, and retrieve relevant passages for systems that generate answers using source material.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

  • Build answers grounded in retrieved documents
  • Add meaning-based search to a knowledge base
  • Find documents with similar content
ruvnet/RuView9.9만2026년 10월 19일 업데이트

browser

무료

Automate browser tasks with agent-browser, using snapshots that label page elements for the agent. Handle login flows, fill and submit forms, extract page information, and capture screenshots through a repeatable navigation and interaction workflow.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

  • Automating login flows
  • Filling and submitting web forms
  • Extracting product names, prices, and links
ruvnet/RuView9.9만2026년 10월 19일 업데이트

ruvnet의 스킬 모두 보기

이 스킬의 문제 신고