hermes-agent-self-evolution README

The project's README from github.com/NousResearch/hermes-agent-self-evolution. As one markdown file: README.md.

🧬 Hermes Agent Self-Evolution

Evolutionary self-improvement for Hermes Agent.

Hermes Agent Self-Evolution uses DSPy + GEPA (Genetic-Pareto Prompt Evolution) to automatically evolve and optimize Hermes Agent's skills, tool descriptions, system prompts, and code — producing measurably better versions through reflective evolutionary search.

No GPU training required. Everything operates via API calls — mutating text, evaluating results, and selecting the best variants. ~$2-10 per optimization run.

How It Works

Read current skill/prompt/tool ──► Generate eval dataset
                                        │
                                        ▼
                                   GEPA Optimizer ◄── Execution traces
                                        │                    ▲
                                        ▼                    │
                                   Candidate variants ──► Evaluate
                                        │
                                   Constraint gates (tests, size limits, benchmarks)
                                        │
                                        ▼
                                   Best variant ──► PR against hermes-agent

GEPA reads execution traces to understand why things fail (not just that they failed), then proposes targeted improvements. ICLR 2026 Oral, MIT licensed.

Quick Start

# Install
git clone https://github.com/NousResearch/hermes-agent-self-evolution.git
cd hermes-agent-self-evolution
pip install -e ".[dev]"

# Point at your hermes-agent repo
export HERMES_AGENT_REPO=~/.hermes/hermes-agent

# Evolve a skill (synthetic eval data)
python -m evolution.skills.evolve_skill \
    --skill github-code-review \
    --iterations 10 \
    --eval-source synthetic

# Or use real session history from Claude Code, Copilot, and Hermes
python -m evolution.skills.evolve_skill \
    --skill github-code-review \
    --iterations 10 \
    --eval-source sessiondb

What It Optimizes

Phase Target Engine Status
Phase 1 Skill files (SKILL.md) DSPy + GEPA ✅ Implemented
Phase 2 Tool descriptions DSPy + GEPA 🔲 Planned
Phase 3 System prompt sections DSPy + GEPA 🔲 Planned
Phase 4 Tool implementation code Darwinian Evolver 🔲 Planned
Phase 5 Continuous improvement loop Automated pipeline 🔲 Planned

Engines

Engine What It Does License
DSPy + GEPA Reflective prompt evolution — reads execution traces, proposes targeted mutations MIT
Darwinian Evolver Code evolution with Git-based organisms AGPL v3 (external CLI only)

Guardrails

Every evolved variant must pass:

  1. Full test suite — pytest tests/ -q must pass 100%
  2. Size limits — Skills ≤15KB, tool descriptions ≤500 chars
  3. Caching compatibility — No mid-conversation changes
  4. Semantic preservation — Must not drift from original purpose
  5. PR review — All changes go through human review, never direct commit

Full Plan

See PLAN.md for the complete architecture, evaluation data strategy, constraints, benchmarks integration, and phased timeline.

License

MIT — © 2026 Nous Research

hermes-agent-self-evolution (NousResearch/hermes-agent-self-evolution)

Searching Database…

5.5K ·GitHub
Hermes-agent-self-evolution
What is this?It is an evolutionary self-improvement system for Hermes Agent that uses DSPy and GEPA (Genetic-Pareto Prompt Evolution) to automatically optimize the agent's skills, tool descriptions, system prompts, and code, producing measurably better versions through reflective evolutionary search.
Is this a general optimization library, or is it built specifically around Hermes Agent?It is purpose-built for Hermes Agent: it operates on Hermes Agent's actual SKILL.md files, tool descriptions, and prompts rather than being a generic optimizer you'd point at any project.
How do I run an optimization pass?The README includes a Quick Start section with the setup commands; the pipeline then mutates text, evaluates results, and selects the best variants automatically via API calls.
What parts of this actually work today?Only Phase 1 is implemented, which evolves skill files (SKILL.md) using DSPy + GEPA; optimizing tool descriptions, system prompt sections, tool code, and a continuous improvement loop are all still planned.
What makes GEPA different from just retrying prompts randomly?GEPA reads execution traces to understand why a variant actually failed, not just that it failed, and then proposes targeted improvements rather than random mutations.
What guardrails stop a bad evolved variant from shipping?Every evolved change must pass the full pytest suite at 100%, respect size limits (skills under 15KB, tool descriptions under 500 chars), stay cache-compatible, preserve the original semantic purpose, and go through human PR review before merging.
Why is this tied to ICLR 2026?The GEPA method underlying the optimization engine was accepted as an ICLR 2026 Oral paper, giving the reflective evolutionary search approach academic validation.
Where can I see the full roadmap?PLAN.md in the repository covers the complete architecture, evaluation data strategy, constraints, benchmark integration, and phased timeline beyond what's in the README.
What are similar projects?imp, rrsi, Proteus, SkillOpt, darwin-skill, wikiskill, awesome-rsi, Raven
Which topics is it in?nous-research, recursive-self-improvement, self-improvement, self-evolving-agents, hermes-agent, agentic, optimization, autonomous-agents, evaluation, prompt-engineering, agent-skills, agent
Creator
NousResearch
Size
<0.1 MB
Stars
5.5K
Updated
CommitJun 17
Age
7mo
Commits · 90 Days