Map the Hermes Agent Ecosystem With a 6k-Star Curated Directory

Hermes Agent, the self-improving open-source agent from Nous Research, has an ecosystem that grows faster than its official docs can catalog. awesome-hermes-agent is a 6k-star, independent directory for exactly that ecosystem: skills, plugins, memory providers, surfaces, and bridges, each carrying a maturity tag so you can tell production tooling from a weekend experiment.

Why This Skill Matters

The directory is explicitly not an official Nous Research project — it is 0xNyk's editorial layer on top, and its independence shows in the mechanics. Every entry carries one of three maturity tags the README defines up front: production (stable, documented, actively maintained), beta (works but evolving), or experimental (proof of concept). The README describes tags as editorial snapshots based on documentation, setup evidence, maintenance, and adoption signals, and tells you to re-check the linked project before depending on it.

A mechanical count puts the list at 253 tagged entries, 137 of them under Skills & Plugins alone. Other sections cover Tools & Utilities, Integrations & Bridges, Memory Providers, Domain Applications, Multi-Agent & Swarms, Forks & Derivatives, Guides & Documentation, and Detection & Media Forensics, plus Operational Playbooks and Level-Up Blueprints that bundle entries into opinionated stacks.

Using the Directory

There is no installer — this is a reading list. The README's own start path has three steps:

1. Run Hermes — the official docs quickstart covers install, CLI, and first chat.
2. Add skills — the directory points at wondelai/skills as a broad production-tagged library, or litprog-skill for literate programming.
3. Add a surface — hermes-workspace for a Hermes GUI, or mission-control for a fleet and dispatch dashboard.

Then work through the full list. Two sections reward a slow read. Check the trust boundary asks, before enabling any community skill or plugin: who can trigger it, which tools it receives, where commands execute, what credentials it can read, and how you can stop it. And the Featured section states its own business model plainly: featured listings are paid advertising placements, not security endorsements.

Real Workflow: Pick a Memory Provider

Memory is where the directory earns its keep. The Memory Providers section alone tags 23 entries, including production picks like mem0 (a universal memory layer with official Hermes integration) and supermemory (cloud or self-hosted memory behind a single API), alongside betas such as Mnemosyne, a local-first, sub-millisecond system built specifically for Hermes with SQLite and sqlite-vec hybrid search.

A practical pass: run stock Hermes memory first, read the trust-boundary checklist, shortlist two providers whose tags and descriptions match your constraint (local-only, self-hosted, or managed), then follow the entry's own README. The Level-Up Blueprints section encodes this exact reasoning as named stacks — "Memory stack that actually compounds" sequences built-in memory, then honcho-self-hosted, then hindsight as your history grows.

Tips

• Treat the maturity tag as a filter, not a guarantee — the README says tags are editorial snapshots and can age.
• The ecosystem status block is dated (last reviewed 2026-07-16, Hermes v0.18.2 at that time), so check Hermes's releases page for the current version.
• wondelai/skills is covered in depth in the wondelai-skills tutorial — the directory tags it production in the agentskills.io ecosystem section.
• The Contributing section accepts entries by issue — the maintainers ask that suggestions relate to Hermes or the agentskills.io standard and have a clear README.
• The list is CC BY 4.0; every entry keeps its own license.

When Not to Use This

This is a directory for the Hermes Agent ecosystem specifically — if you run Claude Code or Codex without Hermes, most entries target a runtime you are not using. It is also a discovery aid, not a vetting service: the README itself says an ecosystem listing is not a security endorsement, so anything you enable still needs the trust-boundary pass.

See the leaderboard for more skills.