Hermes Curator Evolver: plugin for Hermes Agent

Evidence-driven skill evolution plugin for Hermes Agent with dry-run proposals and guarded apply

What Hermes Curator Evolver does

The plugin reads real Hermes session evidence and produces local reports, dry-run proposals and candidate searches for improving skills. Nothing changes silently: writes need explicit approval, an exact target-hash check and a backup, and they can be rolled back. Unattended maintenance is limited to bounded managed blocks. The default path is local and model-free, and semantic search or rerank is optional.

Two commands extend the review workflow. auto-run --variants N generates up to four deterministic, model-free update variants, scores them with local safety and quality signals, and picks a winner; the default is one variant. candidates-mine and candidates-list load redacted session evidence into a SQLite review queue and classify findings as memory, skill_update, skill_new, replay_benchmark or ignore, without writing memory or editing skills. The project says it is inspired by SkillClaw and is not an AutoML system or a skill marketplace.

Key features

Evidence reports built from real Hermes sessions before any skill change
Dry-run proposals that can be reviewed like maintenance notes
auto-run --variants N compares up to four bounded update variants
candidates-mine builds a SQLite review queue of memory, skill and benchmark candidates
Guarded apply with write approval, target-hash checks, backups and rollback
Optional semantic search and rerank

When to use it

Keeping a large skills library from going stale after heavy use
Deciding which session lessons should become memory, a skill patch or a new skill
Running safe daily maintenance on managed blocks without handing the whole library to automation

Who it is for: Hermes Agent users who rely on many skills and want evidence-based, reviewable maintenance of them.

How it fits with Hermes Agent

A Hermes plugin installed with hermes plugins install that works on the skills in your Hermes skills directory.

How to install Hermes Curator Evolver

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

hermes plugins install pingchesu/hermes-curator-evolver --enable
uv pip install --python ~/.hermes/hermes-agent/venv/bin/python -e ~/.hermes/plugins/curator-evolver

FAQ

What is Hermes Curator Evolver?

Hermes Curator Evolver is a Hermes Agent plugin that helps skills improve from real usage. It turns session evidence into reports and dry-run proposals, then applies updates only with review, backups and rollback.

Can Hermes Curator Evolver change my skills automatically?

Not by default. It starts with reports and dry-run proposals, and writes need explicit approval, a target-hash check and a backup. Unattended runs are limited to bounded managed blocks.

How do I install Hermes Curator Evolver?

Run hermes plugins install pingchesu/hermes-curator-evolver --enable, then install the package into the Hermes virtual environment with the uv pip install command from the README.