clawhub/NousResearch/hermes-agent

• **Archetype**: Full-stack AI agent platform with skills, plugins, and learning loop
• **Stars**: 65,657 (GitHub) / Discovered via ClawHub skill author ShawnPana (fork)
• **Last pushed**: 2026-04-12
• **License**: MIT
• **Discovered**: 2026-04-12
• **Source**: clawhub-skills (indirect -- ShawnPana/browser-use ClawHub skill led to hermes-agent fork)
• **Skills found**: 25+ skill directories across 15 categories

Summary

NousResearch's Hermes Agent is a self-improving AI agent with a built-in learning loop, skill system, plugin architecture, multi-platform messaging gateway, and cron scheduler. It is the most feature-rich open-source agent platform found in this research batch.

Key Architecture Components

• **Skills system**: Organized by domain category (software-development, research, creative, data-science, devops, productivity, social-media, etc.) with SKILL.md files containing structured front matter + procedural content
• **Plugin system**: Context engine and memory plugins (mem0, honcho, openviking, byterover, holographic, retaindb, supermemory, hindsight)
• **Learning loop**: Agent creates skills from experience, improves them during use, builds user model across sessions
• **Multi-platform gateway**: Telegram, Discord, Slack, WhatsApp, Signal, Email
• **Cron scheduler**: Natural language scheduled automations with cross-platform delivery
• **Terminal backends**: Local, Docker, SSH, Daytona, Singularity, Modal (serverless)
• **Trajectory compression**: For training next-gen tool-calling models
• **ACP (Agent Communication Protocol)**: Registry and adapter for agent-to-agent communication
• **Batch runner**: Trajectory generation for reinforcement learning
• **Toolset distributions**: Configurable toolset combinations

Notable Skills

1. **software-development/systematic-debugging**: 4-phase root cause investigation with "Iron Law: NO FIXES WITHOUT ROOT CAUSE" -- extremely well structured 2. **software-development/test-driven-development**: Full TDD workflow skill 3. **software-development/writing-plans**: Planning methodology 4. **software-development/subagent-driven-development**: Multi-agent delegation patterns 5. **software-development/requesting-code-review**: Code review workflow 6. **research/research-paper-writing**: End-to-end ML paper pipeline (experiment design -> analysis -> drafting -> revision -> submission) targeting NeurIPS/ICML/ICLR/ACL/AAAI/COLM with statistical analysis, citation verification, and iterative feedback loops 7. **research/arxiv**: arXiv paper discovery and analysis 8. **creative/creative-ideation**: Constraint-driven project generation with rich constraint library 9. **creative/songwriting-and-ai-music**: Music creation workflow 10. **creative/manim-video**: Mathematical animation creation 11. **productivity/linear**: Linear issue tracker integration 12. **autonomous-ai-agents/**: Skills for claude-code, codex, hermes-agent, opencode agent management

Memory Plugins

9 memory backend integrations: mem0, honcho, openviking, byterover, holographic, retaindb, supermemory, hindsight -- each providing different persistence and retrieval strategies.

Assessment

**HIGHEST VALUE** -- This is the single most valuable repository discovered in this batch, and possibly the most valuable external reference repo for babysitter overall. Reasons:

1. **Skills architecture is directly comparable to babysitter's process library**: Same SKILL.md pattern, same domain categorization, same structured front matter with metadata 2. **Learning loop patterns**: Hermes Agent's self-improving skill creation could inform babysitter's retrospect and process improvement workflows 3. **Research paper writing pipeline**: One of the most sophisticated domain-specific skills seen anywhere -- iterative multi-phase workflow with feedback loops, perfect for babysitter process extraction 4. **Systematic debugging methodology**: Already proven and battle-tested, with clear phases and "Iron Law" enforcement 5. **Multi-platform delivery model**: Gateway pattern for Telegram/Discord/Slack could inform babysitter's harness adapter architecture 6. **Trajectory compression**: Token optimization patterns relevant to babysitter's compression layer 7. **Cron scheduler integration**: Natural language cron patterns relevant to babysitter's scheduled orchestration

Extraction Priority

**P0 -- Highest priority extraction target**

Processes

1. **Research Paper Writing Pipeline** (specializations/domains/science/ml-research/): 8-phase iterative pipeline covering literature review, experiment design, execution/monitoring, analysis, drafting, self-review, revision, and submission. Targets NeurIPS/ICML/ICLR/ACL/AAAI/COLM. Includes statistical analysis, citation verification, and feedback loops.

2. **Systematic Debugging Methodology** (methodologies/): 4-phase root cause investigation with strict "no fixes without understanding" enforcement. Directly maps to a babysitter process with breakpoint gates between phases.

3. **Creative Ideation Process** (specializations/domains/creative/): Constraint-driven project generation with rich constraint library across developer, artist, writer, and maker domains.

4. **Subagent-Driven Development** (methodologies/): Multi-agent delegation patterns for parallel workstream execution with coordination protocols.

5. **Self-Improving Skill Creation** (methodologies/): Meta-process for creating new skills from agent experience, improving them during use, and maintaining a skill library.

Plugin Ideas

1. **Hermes Gateway Bridge Plugin**: Enable babysitter to receive work from Telegram/Discord/Slack via Hermes Agent's gateway protocol, extending babysitter's reach beyond CLI harnesses.

2. **Trajectory Compression Plugin**: Adapt Hermes Agent's trajectory_compressor.py for babysitter's run journal compression, reducing storage and improving replay performance.

3. **Memory Backend Abstraction Plugin**: Abstract over multiple memory backends (mem0, honcho, etc.) for babysitter's state persistence, similar to Hermes Agent's plugin/memory architecture.

Library Mapping

Plugin Marketplace Mapping