hermes-subagent

Delegate multi-step tasks to a Hermes Agent sub-agent running in an isolated Python sandbox.

530|67|Updated Mar 13, 2025
One-click install
npx skills add https://github.com/agnt-gg/agnt --skill hermes-subagent
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: hermes-subagent
Source: https://github.com/agnt-gg/agnt/tree/main/backend/skills/hermes-subagent
Command: npx skills add https://github.com/agnt-gg/agnt --skill hermes-subagent

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires hermes-agent, python-dotenv, and includes scripts (resource) and references (resource) components.

What problem does it solve?

Offloading long-running, tool-heavy, or multi-step work to an autonomous sub-agent keeps the main chat context clean, but invoking Nous Research's Hermes Agent correctly is error-prone: wrong constructor arguments, silent model resolution failures, and OpenRouter credit rejections all cause failed delegations. This Skill encodes the verified working invocation pattern so delegation succeeds on the first try.

Core Features & Use Cases

  • Verified invocation helper: A bundled delegate.py script wraps AIAgent with tested defaults (quiet_mode in the constructor, explicit model and provider, max_tokens capped at 4000) and supports flags for model, session id, toolsets, and iteration limits.
  • Delegation decision guidance: A decision table distinguishes tasks suited for the main chat (quick lookups, image generation, AGNT-specific tools) from tasks suited for Hermes (multi-step research, autonomous browsing, heavy code execution, cross-session continuity).
  • Error recovery playbook: Documented fixes for real observed failures including HTTP 402 credit rejections, quiet_mode TypeErrors, empty responses from unresolved models, and wrong-interpreter ImportErrors.
  • Use Case: Ask the agent to research and compare three open-source agent frameworks; the task is delegated to Hermes via the sandbox Python, and the verbatim result returns with model, token, duration, and cost metadata.

Quick Start

Ask the agent to delegate a research or multi-step task to Hermes, for example by saying "have Hermes research the top open-source agent frameworks and summarize their differentiators."

Frequently Asked Questions about hermes-subagent

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I delegate a task to a Hermes Agent sub-agent?

Run the bundled delegate.py script using the absolute path to the sandbox venv's python.exe, passing the task as a quoted argument. The script loads the sandbox .env, constructs AIAgent with verified defaults, and prints Hermes' final response on stdout.

When should I use a sub-agent instead of handling a task in chat?

Delegate to Hermes for multi-step research, long-running autonomous browsing, heavy code execution, or anything exceeding roughly 20 chat turns. Handle quick lookups, single tool calls, image generation, and conversational replies directly in the main chat.

Why does AIAgent.chat() return None after a long pause?

The model was not resolved because model and provider were not set explicitly. Pass both model='anthropic/claude-sonnet-4.5' and provider='openrouter' to the AIAgent constructor; Hermes' default model resolution fails silently.

How do I fix OpenRouter HTTP 402 requires more credits errors?

Pass max_tokens=4000 to the AIAgent constructor or use the --max-tokens flag on delegate.py. Hermes defaults to reserving 64000 output tokens, which OpenRouter's pre-flight credit check rejects on low-balance accounts.

Can Hermes continue a conversation across multiple sessions?

Yes, pass a stable session_id to the AIAgent constructor or use the --session-id flag on delegate.py. Hermes persists session state in its own SQLite store and reloads prior turns automatically when the same id is reused.

Why does importing AIAgent from run_agent fail?

The wrong Python interpreter is being used; the system Python does not have hermes-agent installed. Always invoke the sandbox venv's python.exe by its absolute path rather than relying on python from PATH.