Hermes Agent's skill system is a curated layer on top of the agentskills.io open standard. Skills are reusable, composable agent capabilities — you install one and the agent gains a new ability without you writing any code.
v0.14.0 added nine new optional skills that you can install via hermes skills. It also made huggingface/skills a trusted default tap (#26219), which means community skills published on Hugging Face install with no extra configuration.
This post is a tour: what each new skill does, when it's worth installing, and where to find each in the upstream repo.
How to install any of these
hermes
> /skills
Browse the list, pick one, install. Alternatively:
hermes skills install <skill-name>
Skills are namespaced. The v0.14.0 additions live under the appropriate category in the hub.
---
1. Hyperliquid — perpetuals and spot trading
PR: #23583 (salvage of #1952)
The Hyperliquid skill wraps the Hyperliquid SDK plus the REST API for perpetual and spot trading on the Hyperliquid DEX. The agent can:
- •Quote prices on perp and spot markets
- •Place market or limit orders
- •Read your positions and PnL
- •Monitor a watchlist
When to install: if you trade on Hyperliquid and want the agent to handle order placement or monitoring from chat. Skip if: you don't trade or you only want read-only market data (use Yahoo Finance below).
Warning: this skill can place real orders with real money. Wire it up only on agents you trust completely, and consider scoping the API key to read-only first.
2. Yahoo Finance — market data
PR: #23590
Live equity quotes, options chains, fundamentals (P/E, market cap, etc.), and historical price series. No API key needed — Yahoo Finance has a public endpoint the skill wraps.
When to install: anyone who wants the agent to answer "what did NVDA close at" or "graph the 1-year of AAPL." Skip if: you need real-time intraday data with sub-second freshness — Yahoo Finance has a delay.
Pairs well with cron — schedule a daily market summary, deliver to your messaging gateway.
3. api-testing — REST and GraphQL debug recipes
PR: #23582 (salvages #1800)
A set of debug recipes for hitting REST endpoints and GraphQL servers from the agent. Handles auth (bearer / basic / OAuth), pagination, multipart, structured response parsing. Includes a "diagnose why this 400 happened" mode that walks through common API debugging patterns.
When to install: developers who use Hermes for backend work and want the agent to be competent at API calls. Skip if: you only use Hermes for code-write tasks, not runtime debugging.
4. Unified EVM multi-chain
PR: #25299 (salvages #25291 + #2010, folds in base/)
One skill that covers Ethereum mainnet plus the major L2s (Base, Arbitrum, Optimism, others) under a unified interface. Read balances, simulate transactions, query contract state across chains without juggling per-chain skills.
When to install: if you do anything on EVM chains and the agent should be aware of the cross-chain picture. Skip if: you only use a single chain — install just that chain's skill instead.
5. darwinian-evolver — evolutionary prompt/skill tuning
PR: #26760
This is the most experimental of the nine. The skill runs an evolutionary algorithm over prompts or skill definitions: mutate variants, score them against a fitness function you provide, select the best, repeat. It's a way to find better prompts than you'd write by hand for a specific task.
When to install: power users who want to tune skills or prompts and have a clear fitness function (a test set, a benchmark, a numeric goal). Skip if: you're using Hermes for general chat — evolver shines on narrow, gradable tasks.
6. osint-investigation — OSINT recipes
PR: #26729 (closes #355)
Open-source intelligence recipes for investigating people, domains, and organizations. Looks up DNS history, WHOIS, certificate transparency logs, social media presence, leaked-email databases (Have I Been Pwned). The agent runs the right query for the question you asked.
When to install: security researchers, journalists, sales/recruiting roles, anyone who legitimately needs to compose multiple OSINT sources fast. Skip if: you don't have a legitimate use case — this skill is a stalker multiplier in the wrong hands.
7. pinggy-tunnel — expose local services to the public internet
PR: #26765
Wraps Pinggy to expose a local service on a public URL — useful for testing webhooks, demoing a local app, or letting an agent on a remote machine reach a service running on your laptop.
When to install: anyone building webhook-driven integrations or who needs short-lived public tunnels. Skip if: you already use ngrok / Cloudflare tunnels — pinggy is one option among several.
8. watchers — RSS / HTTP JSON / GitHub polling
PR: #21881
Cron-native polling for things that don't have webhooks. Configure a watcher with a URL (RSS feed, HTTP JSON endpoint, or GitHub repo/branch), a polling interval, and a notification condition. When the condition fires, the watcher delivers a message through whichever platform you configured.
This skill is in the no-agent cron mode — it doesn't invoke a full agent turn per check, just a lightweight polling worker. Important for things you want to check every 60 seconds without paying for an LLM call each time.
When to install: anyone who wants "tell me when X changes" automation. Skip if: every source you care about has a webhook — use that instead.
9. Notion overhaul — Developer Platform refresh
PR: #26612
The Notion skill was rebuilt against the May 2026 Notion Developer Platform refresh, which broke older clients. The new skill handles Notion's data API, blocks, databases, comments — all of it — through the new endpoints.
When to install: if Notion is your team's knowledge base or task tracker, this restores agent integration to current. Skip if: you don't use Notion.
---
The huggingface/skills tap (#26219)
Hugging Face has a community-published skills directory at huggingface.co/skills. v0.14.0 made this a trusted default tap in the Hermes Skills Hub — meaning skills published there install via hermes skills with no extra source-list configuration.
Practical effect: when somebody publishes a useful skill on Hugging Face, it shows up in your hermes skills browse without any plumbing on your end.
Trust caveat: "trusted tap" means "in the default search list," not "every skill there is audited." See the security model post — read the skill before installing it.
How to find what you need
The Skills Hub UI has been getting better with each release. v0.14.0 added richer info panels (#22905) and per-skill pages in the left sidebar (#26646), so browsing the catalog is less of a wall-of-names.
For programmatic access, hermes skills list works from the CLI; pipe it into grep to find what you're looking for.
Where the ecosystem is going
The pattern across these nine: agent capabilities that used to need custom code now ship as one-line installs. Yahoo Finance, OSINT, multi-chain, Hyperliquid — all of these would've been "write a Python wrapper around the API" five years ago. With agentskills.io as the standard and the huggingface/skills tap as a distribution channel, the curve is bending toward "install, don't build."
If you wrote skills for OpenClaw, the migration guide covers how to bring them over. If you want to publish a new skill, the agentskills.io standard is the place to start — the same skill will work in Hermes and any other agent that adopts the standard.
Read more
- •v0.14.0 release notes — new skills section
- •agentskills.io — the open skills standard
- •Hugging Face skills — the community-published skill tap
- •Hermes Agent docs — Skills System
- •The Hermes Agent project on GitHub