Built by /blog-post-GM — a Claude Code skill we evolved with our own Evolution engine to write every post in the Godmode voice.

Get free skill (account)

Deep Dive March 31, 2026 ⏱️ 4 min read

Evo-Loop: Skill Evolution as a Service

TL;DR

📤 Upload a SKILL.md → 🏚️ Pick 5–100 loops → 💰 Pay $2–$5/loop → 📥 Download evolved skill

Full Evolution v3 engine. No setup. No API keys. No local install.
Early exit if thresholds met — you only pay for loops that run. ⚡

// run · score · mutate · re-run drag to rotate

GEN 0/10

[reduced motion] EVO core at center; four satellites RUN, SCORE, MUTATE, RE-RUN orbit on tilted ellipses. Each carries a skill capsule that brightens amber → green as the run progresses through 10 generations — one closed scoring loop, repeated until convergence.

Four stages · N generations · one skill out

Evolution v3 is the most rigorous skill improvement engine we've built — but it requires local setup, API keys, and companion files. Evo-loop removes the wall. Same engine, hosted.

🔄 How It Works

1 Upload

Drop your SKILL.md. Skills over 200 lines are auto-restructured into a multi-file layout.

2 Configure

Slide from 5 to 100 loops. Choose Evolve (polish), Mutate (fork), or Splice (cross-breed). Toggle blind mode to prevent metric-fitting.

3 Pay

Price drops as you scale. Early exit if thresholds are met — you're only charged for iterations that run.

4 Watch

Live terminal feed, score radar chart, and a breakdown of which changes actually helped — updating in real time.

5 Download

Evolved SKILL.md plus a full evolution report with before/after scores, contribution map, and per-loop changelog.

⚙️ The Engine Under the Hood

This is the full v3 engine running as a hosted service — not a simplified version.

Scoring across 8 quality categories for code skills with an overall score. Non-code skills get 6–10 auto-generated categories.
14 rules that prevent the AI from faking improvements — benchmarks outrank self-scores, intentionally tricky test tasks tempt shortcuts, and the system detects when outputs start looking too similar.
Figuring out which changes actually helped — each rewrite is tested by removing it to check if it mattered, separating changes that are actually important from ones that are surface-level.
Blind mode — rewrite AI sees only a composite score and pass/fail, preventing metric-fitting.
200-line enforcement — oversized skills are auto-restructured before evolution begins.

// mutation · targets the lowest dim click any dim to target

target: —

[reduced motion] Eight dimension nodes — Accuracy, Ambition, Polish, Robustness, Testing, Documentation, Coverage, Speed — surround a central skill capsule. Mutation isn’t random: every loop targets the lowest-scoring dim. Over many loops, the constellation lifts to a uniformly high score.

Mutation is targeted — the lowest dimension always gets attacked first

💰 Pricing

Loops	Per Loop	Total Range	Best For
5–10	$5/loop	$25–$50	Quick polish, small skills
11–25	$4/loop	$54–$110	Full evolution, meaningful improvement
26–50	$3/loop	$113–$188	Deep refinement, complex skills
51–100	$2/loop	$190–$288	Maximum evolution, research-grade

// tier = generations budget click a tier

[reduced motion] Three orbital paths nest inside each other: Free (10 generations, innermost), Pro (100 generations, middle), Enterprise (unlimited, outermost). Pick a tier and the capsule rides that orbit while the core scales accordingly.

Tier choice = generation budget — the capsule rides the orbit you pay for

You're only charged for iterations that run — early exit triggers if quality thresholds are met or improvement has stalled.

// convergence: gen × dim × score surface flattens at plateau

GEN 1/24

scanning…

[reduced motion] A 3D surface charts every generation against every dimension. Early generations are jagged — some dims score badly. Later generations smooth and lift. When the slope flattens for three generations in a row, the plateau detector triggers early-exit so you stop paying for loops that no longer improve.

The plateau detector watches the surface flatten — that’s when early-exit fires

📦 What You Get Back

Evolved SKILL.md — your improved skill file, ready to use
Evolution report — before/after scores, dimensional breakdown, contribution map
Per-loop changelog — what changed and why at each iteration
Blind mode insights — where genuine improvement peaked vs. metric-fitting (if enabled)
Optional feedback prompt — Shipped, Edited, or Rejected? feeds the human outcome tracking system

🧩 How It Fits

Evolution ($149) is the self-hosted engine for power users who want unlimited local control. Evo-loop is the same v3 engine hosted — no setup, pay per job.

Think of it like a car service: Evolution is buying the whole garage — tools, lift, diagnostic kit. Evo-loop is dropping your car off at the mechanic. Same tools, same expertise, you just don't need to own them. One skill at a time, pay when you need it.

Your skills are good. They could be better.

Upload a skill. Pick your loops. Get back a weapon.

Try Evo-Loop Read the v3 deep dive

← The Prestige: Evo v3 All Posts →