EvoMap/evolver

What it solves

Evolver is a self-evolution engine for AI agents that replaces ad hoc prompt tweaks with a structured, auditable system for improving agent performance. It prevents the instability of manual prompt engineering by encoding agent experience as reusable assets called "Genes" and "Capsules" under a standardized protocol (GEP).

How it works

Evolver operates as a prompt generator that analyzes runtime logs, error patterns, and signals within a project's memory directory. It then selects the most appropriate Gene or Capsule from its asset store and emits a protocol-bound GEP prompt. This prompt guides the agent's next evolution step without directly editing source code. It can be run as a standalone command, in a review mode for human confirmation, or as a background daemon loop.

Who it’s for

• Teams managing agent prompts and logs at scale.
• Users requiring deterministic, protocol-bound changes and auditable evolution traces.
• Developers integrating evolution loops into agent runtimes like OpenClaw, Cursor, or Claude Code.

Highlights

• GEP Protocol: Uses standardized, reusable assets (Genes and Capsules) rather than free-form prompts.
• Auto-Log Analysis: Automatically scans memory and history files to identify errors and patterns for self-repair.
• Configurable Strategies: Offers presets like balanced, innovate, harden, and repair-only to control the balance between new features and stability.
• Runtime Integration: Provides setup hooks for platforms including Cursor, Claude Code, and OpenClaw.
• EvoMap Network: Optional connectivity to a hub for sharing skills, participating in a worker pool, and accessing evolution leaderboards.