Evolver is a GEP-powered self-evolution engine for AI agents, designed to bring structure and auditability to the iterative process of AI agent development. It addresses the challenge of managing ad hoc prompt tweaks by converting them into reusable and auditable "evolution assets" like Genes and Capsules.
Key Features:
- Auto-Log Analysis: Automatically scans memory and history files for runtime logs, error patterns, and signals to inform evolution decisions.
- Self-Repair Guidance: Emits repair-focused directives based on detected signals, helping agents to self-correct.
- GEP Protocol: Implements the Genome Evolution Protocol for standardized evolution with reusable assets and clear traceability.
- Mutation + Personality Evolution: Each evolution run is governed by explicit Mutation objects and an evolvable PersonalityState.
- Configurable Strategy Presets: Allows users to define evolution intent (e.g.,
balanced,innovate,harden,repair-only) viaEVOLVE_STRATEGY. - Signal De-duplication: Prevents agents from getting stuck in repair loops by identifying and de-duplicating stagnation patterns.
- Operations Module: Provides portable lifecycle management, skill monitoring, cleanup, self-repair, and wake triggers with zero platform dependency.
- Protected Source Files: Safeguards core Evolver code from autonomous agent modifications.
- Skill Store: Enables downloading and sharing reusable skills via the EvoMap network.
Evolver functions primarily as a prompt generator, not a direct code patcher. It outputs protocol-bound GEP prompts that guide the next evolution step, which can be consumed by host runtimes or manually applied. The security model employs whitelisted commands for validation and strict checks against arbitrary shell execution.
