Let Skills Evolve Collectively with Agentic Evolver
WikiSkill (arXiv:2608.27454) for Hermes Agent — self-evolving agent skills via a persistent knowledge wiki. Faithful Algorithm 1 implementation with real agent runs, isolated skill gating, and a documented live run log.
Evidence-driven skill evolution for Hermes Agent — reports, dry-run proposals, candidate search, and guarded apply
Flow-driven recursive skill evolution for agentic LLM orchestration.
Pulse — Hermes-style self-improving AI agent. Reliability-first rebuild with evaluated skill self-evolution, multi-agent team orchestration, dialectic user modeling, and fully self-hosted default stack (Ollama + SQLite FTS5). Apache 2.0 license.
Multi-agent framework for Claude Code: tier-aware acceptance (S/M/L), acceptor/optimizer split, filesystem-isolated adversary review, cross-family second-opinion via Codex MCP, SkillOpt-style skill evolution, event ledger observability, human as supreme judge
SkillTV-Bench: a benchmark for evaluating, evolving, and analyzing judge skills on agent trajectories.
Bionic memory for AI agents: GraphRAG, predictive memory, dream consolidation, governed skill evolution, MCP, REST API, and dashboard.
技能自进化系统 v2.0.0 — 8阶段Loop + 跨模型双裁判 + 30题测试集
数字员工操作系统 — AI agent 像真正员工一样工作、学习、成长的协作平台。3D 虚拟办公室 + 技能进化 + 职业发展 + 记忆系统
MetaClaw fork: self-hosted GPU training (GRPO + LoRA + vLLM), per-agent skill isolation, multi-agent mode routing, training engine bug fixes. Train your AI agent on your own A100s. No cloud dependency.
Zero-dependency AI agent orchestration engine. Parallel DAG execution, skill genetics, ontology routing, human approval gates. 8.7K LOC, <1MB. The embeddable agent engine that learns.
Evolve AI agent skills through chat interactions that persist across devices and users with a collective learning framework.
Human-gated, local-first self-evolution for Codex Skills.
Turn Codex work records into a personal wiki and reusable skill candidates, with source evidence and evaluation gates.
面向企业数据分析与运维诊断的 Agent Harness，统一管理任务规划、工具执行、权限审批、上下文与预算。将执行轨迹提炼为经验和 Skill，通过离线评测、人工审批与版本回滚，实现可追溯的受控方法演进。
SoulSkill packages native AISP skill packages as a local skill library plus Codex and Claude Code plugin adapters. It gives agents and developers a self-contained way to discover, inspect, validate, hash, and distribute AISP skills.
Installable Claude Code plugin for building agent skills that compound: append-only run archives, a model-maintained wiki, an eval gate before any update ships, cross-skill discovery, and an opt-in git sync layer that auto-archives and backs up skill changes.
A privacy-clean framework for evolving, routing, validating, and sharing Codex skills.
To associate your repository with the skill-evolution topic, visit your repo's landing page and select "manage topics."