New: 97,000+ skills indexed from GitHub repos with 100+ stars · every skill safety-checkedBrowse top skillsTop skills
Esc to closeBrowse all skills
Submit a skillBrowse skills
Submit a skillBrowse skills

Agent skill

Darwin Mode Harness Evolution

by ruvnet in ruvnet/RuView

Runs Darwin Mode on an agent harness: mutates one policy file per generation in a sandbox, scores each variant against your tests, and archives only variants that measurably improve.

MITAuto-check passedAgent Workflows

Install Darwin Mode Harness Evolution

skills CLI
$ npx skills add ruvnet/RuView --skill evolve -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install ruvnet/RuView evolve --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/ruvnet/RuView.git skills-src && mkdir -p .claude/skills && cp -r skills-src/harness/wifi-densepose-sar/.claude/skills/evolve .claude/skills/evolve && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
evolve
GitHub stars
97k
Token cost
~693 tokens
SKILL.md length
307 words
Files
1
Skills in repo
24
Repo updated
First seen
Licence
MIT

At a glance

Runs Darwin Mode on an agent harness: mutates one policy file per generation in a sandbox, scores each variant against your tests, and archives only variants that measurably improve.

  • Works in 6 steps: Closed-loop repair is the #1 lever… → Cheap-first + cost-aware routing. Track… → Tier the models (Barbarian & Scholar).… → …
  • Letting a harness improve its own planner or retry policy under test
  • SKILL.md covers Run it, Safety (secure by default) and What the benchmarks taught us…
  • Calls npm and npx

What it does

In this skill the model stays frozen and the harness evolves. Each generation mutates one of seven surface files (planner, context builder, reviewer, and the retry, tool, memory and score policies), runs each child in a sandbox, scores it, and keeps only variants that measurably improve, building an archive of successful descendants. It runs with npm run evolve, which executes your test command for every variant using a deterministic mutator that needs no API key or network, or with npm run evolve:dry, a fast offline mock that runs no tests, or directly through npx metaharness-darwin with sandbox, generations and children options.

Safety is the default: every mutation passes a validateGeneratedCode gate that allows no new imports, network, filesystem, shell, environment access or dependencies, so changes are pure refactoring or tuning, and nothing is promoted without measured improvement. The skill also records lessons from a full SWE-bench Lite run: feeding test failures back and retrying took the resolve rate from 7.7% to 15.3% on the same cheap model, cost per resolved task matters more than resolve rate alone, tiering cheap and frontier models cut cost roughly 6x, and the output-format contract belongs in a system message with an example.

When your agent uses it

  • Letting a harness improve its own planner or retry policy under test
  • Running a safe offline dry run of an evolution cycle
  • Choosing defaults informed by SWE-bench Lite findings

Example prompts

  • “Run npm run evolve for this harness and report which variants were archived.”
  • “Do a dry run of the evolution loop offline first.”
  • “Evolve the retry policy for three generations with four children each.”

Requirements

  • npm, with a test command that can score each variant

Workflow steps

6 steps, taken from the first numbered list in SKILL.md.

  1. Closed-loop repair is the #1 lever (~2×). Feeding test/compiler failure back and retrying took
  2. Cheap-first + cost-aware routing. Track $/resolve, not just resolve-rate; a cheap model
  3. Tier the models (Barbarian & Scholar). Cheap sweep + frontier on only the residual = 33.3%
  4. Put the output-format contract in a system message + example, and size prompts to the model's
  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

What it can do on your machine

Read from SKILL.md and the folder at commit 0ef6b96. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Shell commands in SKILL.md call:

    • npm
    • npx

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md. Its commands use npm and npx, which can reach the network depending on how they are called.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Darwin Mode Harness Evolution loads about 693 tokens when it runs. Until then it costs about 27 tokens; SKILL.md has 307 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~27
When it runs · the whole SKILL.md, loaded when a task matches
~693

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from ruvnet/RuView at commit 0ef6b96, republished under its MIT licence (© ruvnet). 307 words, ~693 tokens.

Download SKILL.mdSave it as .claude/skills/evolve/SKILL.md (or your agent's skills folder).
name
evolve
description
Evolve this harness with Darwin Mode — frozen model, evolving harness (real, sandboxed, safety-gated).

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

bash
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:

bash
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.

© ruvnet, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in harness/wifi-densepose-sar/.claude/skills/evolve of ruvnet/RuView.

Open the folder on GitHubat commit 0ef6b96

Used in 1 other repository

We found 1 copy of this SKILL.md (exact, near-identical or edited) in other folders. This page covers the copy in ruvnet/RuView, which our catalogue first saw on October 7, 2026.

Compare with similar skills

Darwin Mode Harness Evolution next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Darwin Mode Harness Evolution compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Darwin Mode Harness Evolution this skillruvnet/RuView97k—~693Automated safety check: PassMIT
CodeGraph Agent Evalcolbymchenry/codegraph73k—~950Automated safety check: PassMIT
Babysitter Process Runnera5c-ai/babysitter1.8k1 repos~726Automated safety check: NotesMIT
Ax Agent Rlmdosco/aithy107—~8.2kAutomated safety check: PassApache-2.0
OpenHarness End-to-End EvalsHKUDS/OpenHarness16k1 repos~2.1kAutomated safety check: NotesMIT
A-Evolve Agent EvolutionOrchestra-Research/AI-Research-SKILLs13k1 repos~3.6kAutomated safety check: PassMIT

Similar skills

  • CodeGraph Agent Eval

    colbymchenry/codegraph

    Benchmarks how much CodeGraph helps a coding agent on a real repository, comparing runs with and without it for a chosen local or published version.

    73k GitHub stars~950 tokensUpdated today
    Agent WorkflowsAuto-check passed

  • Babysitter Process Runner

    a5c-ai/babysitter

    Execute via @babysitter. Use this skill when asked to babysit a task, do anything that is structured process-driven (even a loop) or whenever it is called…

    1.8k GitHub starsUsed in 1 repo~726 tokens
    Agent WorkflowsAuto-check: notes

  • Ax Agent Rlm

    dosco/aithy

    This skill helps an LLM generate correct AxAgent RLM/runtime code using @ax-llm/ax.

    107 GitHub stars~8.2k tokensUpdated 1 mo ago
    Agent WorkflowsAuto-check passed

  • OpenHarness End-to-End Evals

    HKUDS/OpenHarness

    Validates OpenHarness features by running real multi-turn agent loops with live LLM calls against an unfamiliar codebase, checking actual tool execution.

    16k GitHub starsUsed in 1 repo~2.1k tokens
    Testing & QAAuto-check: notes

  • A-Evolve Agent Evolution

    Orchestra-Research/AI-Research-SKILLs

    Guidance for using A-Evolve to improve an AI agent automatically, evolving its prompts, skills and memory against a benchmark through solve, observe and evolve cycles.

    13k GitHub starsUsed in 1 repo~3.6k tokens
    AI & LLM EngineeringAuto-check passed

  • MCP Server Builder

    anthropics/skills

    Official

    Guides the design and implementation of Model Context Protocol servers in TypeScript or Python, from tool naming and error messages to evaluation.

    180k GitHub starsUsed in 62 repos~2.3k tokens
    Agent WorkflowsAuto-check passed

More from ruvnet/RuView

All 24 skills in this repo

  • RuView Advanced Sensing

    ruvnet/RuView

    Reference for RuView's research-grade WiFi sensing features: multistatic fusion, cross-viewpoint geometry, persistent field models, RF tomography, intention signals and mesh security.

    97k GitHub stars~1.2k tokensUpdated today
    Auto-check: notes

  • RuView Sensing Applications

    ruvnet/RuView

    Runs RuView's WiFi sensing applications: presence, vital signs, activity and fall detection, pose estimation, sleep monitoring and environment mapping.

    97k GitHub stars~1.1k tokensUpdated today
    Auto-check: notes

  • RuView CLI, API and WASM

    ruvnet/RuView

    Covers the RuView `wifi-densepose` command line binary, its Axum REST API and the WebAssembly builds for browsers and ESP32, for embedding or scripting RuView.

    97k GitHub stars~1.2k tokensUpdated today
    Auto-check: notes

  • RuView Deployment Configuration

    ruvnet/RuView

    Tunes a deployed RuView WiFi-sensing system without changing code: firmware sdkconfig variants, NVS provisioning over serial, channel and MAC filtering, edge processing tiers and mesh slotting.

    97k GitHub stars~1.7k tokensUpdated today
    Auto-check: notes

  • Agent-Browser Automation

    ruvnet/RuView

    Drives a web browser through the agent-browser CLI, using compact accessibility snapshots with element refs in place of the full DOM to keep context small.

    97k GitHub starsUsed in 4 repos~1.3k tokens
    Auto-check passed

  • RuView Hardware Setup

    ruvnet/RuView

    Brings a RuView CSI sensing node online by building ESP32-S3 or ESP32-C6 firmware, flashing the board, provisioning WiFi and checking the serial output.

    97k GitHub stars~1.8k tokensUpdated today
    Auto-check: notes

Works with

Categories

Questions about Darwin Mode Harness Evolution

What does Darwin Mode Harness Evolution do?

Runs Darwin Mode on an agent harness: mutates one policy file per generation in a sandbox, scores each variant against your tests, and archives only variants that measurably improve. In this skill the model stays frozen and the harness evolves. Each generation mutates one of seven surface files (planner, context builder, reviewer, and the retry, tool, memory and score policies), runs each child in a sandbox, scores it, and keeps only variants that measurably improve, building an archive of successful descendants.

When should I use Darwin Mode Harness Evolution?

Darwin Mode Harness Evolution fits situations like: letting a harness improve its own planner or retry policy under test; running a safe offline dry run of an evolution cycle; choosing defaults informed by SWE-bench Lite findings.

How do I install Darwin Mode Harness Evolution in Claude Code?

Run `npx skills add ruvnet/RuView --skill evolve -a claude-code`. Or copy the skill folder (harness/wifi-densepose-sar/.claude/skills/evolve in ruvnet/RuView) into .claude/skills/evolve in your project. Claude Code loads it when a task matches its description.

How do I install Darwin Mode Harness Evolution in Codex?

Run `npx skills add ruvnet/RuView --skill evolve -a codex`. Or copy the skill folder (harness/wifi-densepose-sar/.claude/skills/evolve in ruvnet/RuView) into .agents/skills/evolve in your project. Codex loads it when a task matches its description.

Can I use Darwin Mode Harness Evolution in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add ruvnet/RuView --skill evolve -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/evolve, .gemini/skills/evolve, .github/skills/evolve and .opencode/skills/evolve in your project.

What does Darwin Mode Harness Evolution need to run?

Going by SKILL.md and its folder, Darwin Mode Harness Evolution needs the command-line tools its instructions call (npm and npx). Our summary lists: npm, with a test command that can score each variant.

Does Darwin Mode Harness Evolution access the network?

SKILL.md contains no URLs. Its commands use npm and npx, which can reach the network depending on how they are called. This is read from the text; nothing was executed.

Is Darwin Mode Harness Evolution safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Darwin Mode Harness Evolution use?

Darwin Mode Harness Evolution is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Darwin Mode Harness Evolution use?

About 693 tokens (SKILL.md is roughly 2.8k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Darwin Mode Harness Evolution?

Skills that share tags, products or a category with Darwin Mode Harness Evolution: CodeGraph Agent Eval (colbymchenry/codegraph, 73k stars), Babysitter Process Runner (a5c-ai/babysitter, 1.8k stars), Ax Agent Rlm (dosco/aithy, 107 stars) and OpenHarness End-to-End Evals (HKUDS/OpenHarness, 16k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Darwin Mode Harness Evolution?

ruvnet (a GitHub user) maintains it in ruvnet/RuView, which has 96,741 GitHub stars. The repository holds 24 skills in this directory. The repository was last updated on October 7, 2026.

Source: ruvnet/RuView on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.