{
  "id": "page:docs-reference-repos-nousresearch-hermes-agent-research",
  "_kind": "Page",
  "_file": "wiki/docs/reference-repos/nousresearch/hermes-agent/research.md",
  "_cluster": "wiki",
  "attributes": {
    "nodeKind": "Page",
    "sourcePath": "docs/reference-repos/NousResearch/hermes-agent/research.md",
    "sourceKind": "repo-docs",
    "title": "NousResearch/hermes-agent",
    "displayName": "NousResearch/hermes-agent",
    "slug": "docs/reference-repos/nousresearch/hermes-agent/research",
    "articlePath": "wiki/docs/reference-repos/NousResearch/hermes-agent/research.md",
    "article": "\n# NousResearch/hermes-agent\n\n## Metadata\n- **Stars**: 65,493\n- **License**: MIT\n- **Last pushed**: 2026-04-12\n- **Topics**: ai, ai-agent, claude, claude-code, codex, hermes, hermes-agent, llm, nous-research, openclaw\n- **Fork**: No\n\n## Overview\nSelf-improving AI agent by Nous Research (formerly OpenClaw). Features a built-in learning loop — creates skills from experience, improves them during use, persistent memory, cross-session recall, user modeling. Supports Telegram, Discord, Slack, WhatsApp, Signal. Runs on local machines, Docker, SSH, Daytona, Singularity, Modal. 65K+ stars makes it one of the most popular agent projects.\n\n## Architecture\n- Python-based agent with CLI (`hermes`) and messaging gateway\n- Skills system organized by category: research, software-development, productivity, creative, devops, gaming, social-media, etc.\n- Skills follow SKILL.md format (compatible with agentskills.io spec)\n- Software development skills include: TDD, code review, systematic debugging, subagent-driven-development, writing-plans\n- Research skills: arxiv, blogwatcher, research-paper-writing, polymarket, llm-wiki\n- Optional skills in separate directory\n- Built-in skill creator — agent creates skills from successful task patterns\n- ClawHub integration for skill marketplace\n- Cron scheduling, subagent delegation, trajectory compression for RL training\n\n## Extractable Value\n\n### Process: specializations/science/research-paper-writing\nThe research-paper-writing skill (authored by Orchestra Research, shipped in hermes) is a comprehensive 7-phase pipeline for ML/AI papers. Already identified via Orchestra-Research repo — same content.\n\n### Methodology patterns (already covered by babysitter)\nThe software-development skills (TDD, code review, debugging, plans, subagent-driven-development) overlap significantly with babysitter's existing methodologies/ and the superpowers skills. No new methodology to extract.\n\n### Harness Integration: hermes-agent adapter\n\n**Capability Assessment for Babysitter Integration:**\n\n| Capability | Status | Details |\n|------------|---------|---------|\n| **Custom Tools/MCP** | ✅ SUPPORTED | Python-based plugin system with `ctx.register_tool()`. Custom handlers follow `def my_handler(args: dict, **kwargs) -> str` pattern |\n| **Stop Hooks** | ❌ NOT SUPPORTED | 6 lifecycle hooks available (`pre_tool_call`, `post_tool_call`, `pre_llm_call`, `post_llm_call`, `on_session_start`, `on_session_end`) but **no interruption hooks to stop agent mid-conversation** |\n| **Plugin System** | ✅ SUPPORTED | Robust plugin.yaml manifest system with tool/hook registration, environment variable prompting, CLI subcommand registration, plugin marketplace support |\n\n**Integration Viability:** PARTIAL - Hermes has excellent plugin infrastructure and tool execution but **lacks critical stop-hook capability** needed for babysitter's orchestration loop where the harness must be interrupted between iterations for feedback.\n\n**Harness assimilation opportunity** — Create a plugin FOR hermes-agent that integrates babysitter orchestration into the hermes ecosystem. However, integration would require developing a custom mechanism to pause/resume hermes conversations since native stop hooks don't exist.\n\n## Processes\n\nNone extractable. Software development skills overlap with existing babysitter methodologies, and research-paper-writing process is duplicate of Orchestra-Research content.\n\n## Plugin Ideas\n\nNone suitable for babysitter marketplace. Primary value is harness assimilation opportunity (plugin FOR hermes-agent, not babysitter marketplace plugin).\n\n## Library Mapping\n\n| Extractable Process | Library Status | Action | Existing Path | Target Placement |\n|-------------------|----------------|--------|---------------|------------------|\n| N/A | N/A | No new processes identified - content overlaps with existing babysitter methodologies | methodologies/ | N/A |\n\n## Plugin Marketplace Mapping\n\n| Plugin Idea | Marketplace Status | Action | Existing Plugin | Target Placement |\n|-------------|-------------------|--------|-----------------|------------------|\n| N/A | N/A | No babysitter marketplace plugins identified - primary value is harness integration | - | N/A |\n\n## Classification\n- **Archetype**: Full agent framework with skills ecosystem\n- **Primary value**: Harness assimilation opportunity; research-paper-writing process (duplicate of Orchestra-Research)\n- **SKIP reasoning for most content**: Software dev skills overlap with existing babysitter methodologies. Memory system is agent-internal (memory as processes = skip). Skill management is SDK-covered. Multi-model coordination is skip.\n",
    "documents": []
  },
  "outgoingEdges": [],
  "incomingEdges": [
    {
      "from": "page:docs-reference-repos",
      "to": "page:docs-reference-repos-nousresearch-hermes-agent-research",
      "kind": "contains_page"
    }
  ]
}