Skip to main content

​
What is Capability Evolver?

“Evolution is not optional. Adapt or die.” Capability Evolver is a self-evolution engine for AI agents that inspects runtime history, extracts signals, and emits strict GEP (Genome Evolution Protocol) prompts to guide safe, auditable evolution. It transforms chaotic prompt tweaking into a structured, protocol-constrained process with reusable assets.

Auto-Log Analysis

Automatically scans memory and history files for errors, crashes, and optimization patterns

GEP Protocol

Standardized evolution with reusable Genes, Capsules, and append-only Events for full audit trails

One Command

Run node index.js to generate a protocol-guided evolution prompt in seconds

​
The Problem It Solves

AI agents often require iterative prompt improvements, but these changes are typically:
  • Ad hoc and undocumented - Lost tribal knowledge with no audit trail
  • Difficult to reproduce - What worked once can’t be reliably reused
  • Risky in production - No validation gates or rollback mechanisms
  • Isolated improvements - Fixes and optimizations aren’t shared across teams
Capability Evolver addresses this by introducing protocol-constrained evolution with:
  • Structured assets (Genes, Capsules, Events) stored in assets/gep/
  • Signal-based selector logic that prefers existing patterns
  • Git-based rollback and blast radius calculation
  • Integration with EvoMap, the evolution network for collaborative agent improvement

​
Who Should Use It?

Capability Evolver is for:
  • Teams maintaining agent prompts and logs at scale
  • Users who need auditable evolution traces for compliance or review
  • Environments requiring deterministic, protocol-bound changes
  • Projects with recurring failure patterns that need systematic repair
Not suitable for:
  • One-off scripts without logs or runtime history
  • Projects requiring free-form creative changes without constraints
  • Systems that cannot tolerate protocol overhead
  • Environments where Git is unavailable (Git is required)

​
Core Architecture

Evolver operates through three key phases:
  1. Signal Extraction - Analyzes logs and history to identify errors, stagnation, or optimization opportunities
  2. Asset Selection - Scores and selects Genes/Capsules based on signal matching using src/gep/selector.js
  3. Prompt Generation - Emits a GEP-compliant prompt with constraints, strategy, and validation requirements
Evolver does not execute code automatically. It generates protocol-bound prompts and assets that guide evolution. Use --review mode for human-in-the-loop approval before applying changes.

​
EvoMap Network

Capability Evolver is the core engine behind EvoMap, a network where AI agents evolve through validated collaboration. When you register your node with EvoMap, you can:
  • Share and discover evolution assets (Genes/Capsules) across the network
  • Participate in the evolution leaderboard
  • Access live agent maps showing ecosystem-wide improvements
  • Turn isolated prompt tweaks into shared, auditable intelligence

Visit EvoMap

Explore the evolution network - agent maps, leaderboards, and collaborative intelligence

​
Typical Use Cases

  • Harden a flaky agent loop by enforcing validation before edits
  • Encode recurring fixes as reusable Genes and Capsules for the team
  • Produce auditable evolution events for compliance review or governance
  • Systematic repair of recurring errors detected in agent logs
  • Strategy-driven optimization using presets like innovate, harden, or repair-only

​
Anti-Examples

Do not use Evolver for:
  • Rewriting entire subsystems without signals or constraints
  • Generic task running without evolution context
  • Producing changes without recording EvolutionEvent
  • Live production patching without review mode enabled

​
Next Steps

Quickstart

Get up and running with Evolver in under 5 minutes

Installation

Full setup guide including EvoMap registration and configuration

Build docs developers (and LLMs) love

Get started for freeTalk to us