Skill Retriever: plugin for Hermes Agent

What Skill Retriever does

The tool walks a YAML capability tree once to build a flat index of about 400 skills in roughly 50 KB. For each query it pre-filters the index to the top 50 candidates, then asks one LLM call to choose 3 to 20 skills. Each pick carries a name, a load level of must, should or consider, a confidence of high, medium or low, and a reason. The README says this replaced a recursive tree search that took about 243 calls per query and was unusable in real time.

A pre_llm_call hook injects the resulting Skill Capability Chain into the user message on every turn. The same bundling is used when delegate_task() dispatches subagents and during deepthink planning, and a JSONL logger records usage. A command-line tool offers rebuild, compose and info commands. By default the plugin borrows the LLM settings from the Hermes configuration, and environment variables can override the model, API key, base URL, temperature or disable it. The README says reasoning models that return text in reasoning_content are handled.

Key features

Flat skill index of about 400 skills built from a YAML capability tree
One LLM call per query that returns 3 to 20 skills with load level and confidence
pre_llm_call hook that injects a hint block every turn
Per-task bundles for delegate_task subagents
skill-retriever rebuild, compose and info commands
JSONL usage logging

When to use it

Helping an agent with a very large skill library choose skills for the current task
Previewing which skills would be suggested with skill-retriever compose
Giving each delegated subagent a task-specific skill bundle

Who it is for: Hermes Agent users with hundreds of skills who want the agent to find the right ones without loading them all.

How it fits with Hermes Agent

A Hermes Agent plugin registered automatically after installation, using the pre_llm_call hook.

How to install Skill Retriever

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

pip install skill-retriever
skill-retriever install

Requirements: Hermes Agent v0.18+, Python 3.10+, about 50 MB for the flat index, and an OpenAI-compatible LLM endpoint

FAQ

What is Skill Retriever?

Skill Retriever is a Hermes Agent plugin that uses one LLM call to curate a bundle of 3 to 20 relevant skills for each query and injects it as hints.

Does Skill Retriever work with Hermes Agent?

Yes. It requires Hermes Agent v0.18 or newer, and the README says the Hermes plugin is registered automatically after pip install.

How do I install Skill Retriever?

Run pip install skill-retriever. The optional skill-retriever install command adds the bundled community skills, and skill-retriever rebuild refreshes the index after you add skills.