Live Curated Resource Directory

Best Hermes Skills in 2026:
The 17 I Would Install First

A raw install is smart. A skilled install is unfair. Explore the verified 17-skill modular setup with live GitHub star counts, copyable commands, day-one installation order, and context budget rules.

17

Skills Analyzed

482K+

Verified Stars

05

Day-One Stack

$0

Total Tooling Cost

Best Hermes Skills 2026: seventeen agent skill modules loading into a single runtime
Executive Summary // 30-Second Answer

If I wiped Hermes today, I would not open a blank chat and figure it out. I would install skills first. The seventeen below cover almost every real job I hand an agent: research, engineering, security, video, browser control, writing cleanup, repo memory, and loop discipline.

The 5 Day-One Pillars: Agent-Reach (web access), addyosmani/agent-skills (engineering gates), Humanizer (writing cleanup), Defuddle (clean web input), and Minions (parallel task mission control).
Live Verification Audit

By Muhammad Shadab Shams

AI Automation Consultant running production agent clusters across live properties. Every GitHub star count, owner, and install path was re-verified live via the GitHub API on August 27, 2026.

Verified Repos: 14/173 Flagged Unverified
Day-One Architecture

Install These Five on Day One

Loading all 17 skills immediately will cause you to spend your first week debugging your setup and blowing your context window. These five form the optimal zero-cost stack covering input, process, and output.

01
Engineering Discipline

addyosmani/agent-skills

Production engineering lifecycle skills (define, plan, build, verify, review, ship) that force senior developer habits onto coding agents.

02
Public Web Access

Agent-Reach

A CLI and skill pack giving your agent direct read access to the public internet across X, Reddit, YouTube, GitHub, Bilibili, and XiaoHongShu with zero paid API keys.

03
De-slopping & Voice Matching

Humanizer

De-slop writing skill that strips recognizable AI tells (hype adjectives, uniform lists, structural cliches) based on Wikipedia's AI-detection markers.

04
Clean Web Input

Defuddle

Clean web content extractor by Kepano (Obsidian CEO). Takes messy web URLs and returns pure article Markdown, stripping ads, cookies, and boilerplate.

05
Parallel Mission Control

Minions

Kanban mission control dashboard for Hermes and autonomous agents. Live tool call streaming, human sign-off gates, and local SQLite persistence.

Interactive Skill Matrix

Explore All 17 Hermes Skills

#01
238,275
Engineering & Shipping

Matt Pocock skills

mattpocock/skills

The most-starred skill pack on GitHub by a wide margin, built on small, composable, editable skills rather than one giant opaque framework.

What it fixes:Big process frameworks take control away from you. When your agent has a giant opaque process and something breaks, you cannot debug it. When it has no process at all, work drifts. Matt's set sits deliberately in the middle: small enough that you can open one and hack it.
TypeScript & app engineeringTicket triageFeature planningDay-to-day shipping
npx skills@latest add mattpocock/skills
GitHub Repository
#02
Day-One90,115
Engineering & Shipping

addyosmani/agent-skills

addyosmani/agent-skills

Production engineering lifecycle skills (define, plan, build, verify, review, ship) that force senior developer habits onto coding agents.

What it fixes:Stops 'vibe coding' with zero validation gates. These skills force habits senior engineers already have: write the spec, cut the work small, test it, review it.
Real software productsFrontend & UI workMulti-day feature engineeringCode review automation
npx skills add addyosmani/agent-skills
GitHub Repository
#03
Day-One75,763
Internet & Research

Agent-Reach

Panniantong/Agent-Reach

A CLI and skill pack giving your agent direct read access to the public internet across X, Reddit, YouTube, GitHub, Bilibili, and XiaoHongShu with zero paid API keys.

What it fixes:Agents hit walls when searching current social signals, technical threads, and tutorials: blocked scrapers, paywalled APIs, and login walls. Agent-Reach bridges this cleanly.
Live content researchCompetitor auditsSocial listeningDebugging community bugs
npx skills add Panniantong/Agent-Reach
GitHub Repository
#04
Skip Initially51,586
Media & Specialists

OpenMontage

calesthio/OpenMontage

An open-source agentic video production system packing 12 production pipelines, 100+ tools, and over 700 skill and knowledge files.

What it fixes:Video production normally fractures across scripts, b-roll, timeline cuts, captions, and encoding. OpenMontage turns your agent into an end-to-end video crew.
Short-form videoProduct explainer videosDemo clip generation
npx skills add calesthio/OpenMontage
GitHub Repository
#05
40,809
Engineering & Shipping

codebase-memory-mcp

DeusData/codebase-memory-mcp

High-speed MCP server that indexes your repository into a persistent knowledge graph across 158 languages with sub-millisecond queries.

What it fixes:Context exhaustion. Reading raw code files over and over burns 80% of your token window. A knowledge graph answers structural queries (call graphs, imports, broken references) without loading full files.
Large production reposMonoreposToken cost reduction (up to 99%)
npx skills add DeusData/codebase-memory-mcp
GitHub Repository
#06
Day-One38,179
Writing & Content

Humanizer

blader/humanizer

De-slop writing skill that strips recognizable AI tells (hype adjectives, uniform lists, structural cliches) based on Wikipedia's AI-detection markers.

What it fixes:Standard AI output sounds like a canned PR release. Humanizer rewrites without preserving the robotic syntax, scans for residual markers, and retains 100% of facts.
Public articles & blogsDocumentationClient emailsLaunch announcements
npx skills add blader/humanizer
GitHub Repository
#07
Skip Initially31,306
Media & Specialists

Anthropic Cybersecurity Skills

mukul975/Anthropic-Cybersecurity-Skills

817 structured cybersecurity playbooks across 29 security domains, mapped to MITRE ATT&CK, NIST CSF 2.0, D3FEND, and NIST AI RMF.

What it fixes:Agents lack framework-aligned security rigor by default. This equips the agent for threat modeling, forensic investigations, and authorized red-team prep.
Authorized security reviewsIncident response draftsThreat modelingSecurity education
npx skills add mukul975/Anthropic-Cybersecurity-Skills
GitHub Repository
#08
Day-One9,173
Internet & Research

Defuddle

kepano/defuddle

Clean web content extractor by Kepano (Obsidian CEO). Takes messy web URLs and returns pure article Markdown, stripping ads, cookies, and boilerplate.

What it fixes:Agents choke on raw HTML. Cookie banners, complex nav trees, and inline scripts burn context and distract models. Defuddle extracts pure signal.
Research pipelinesWeb clippingContext preprocessingPairing with Agent-Reach
npx skills add kepano/defuddle
GitHub Repository
#09
3,024
Engineering & Shipping

make-interfaces-feel-better

jakubkrehel/make-interfaces-feel-better

Specialized UI polish skill focusing on micro-interactions, optical alignment, typography scales, hit areas, hover states, and elevation shadows.

What it fixes:AI-generated interfaces usually work but feel cheap or generic. This skill performs a dedicated sensory pass on UI code to make it feel intentional.
Landing pagesWeb app dashboardsComponent librariesMobile touch targets
npx skills add jakubkrehel/make-interfaces-feel-better
GitHub Repository
#10
Skip Initially2,517
Memory & Self-Improvement

SkillClaw

AMAP-ML/SkillClaw

Evolutionary agent skill system that digests past sessions, refines procedures you actually use, and prunes unused skills automatically.

What it fixes:Static procedures stagnate. SkillClaw turns agent usage history into a self-optimizing feedback loop that gets sharper over time.
Long-term agent setupsContinuous developer optimizationPruning dead skills
npx skills add AMAP-ML/SkillClaw
GitHub Repository
#11
Day-One624
Orchestration & Control

Minions

agent37-platform/minions

Kanban mission control dashboard for Hermes and autonomous agents. Live tool call streaming, human sign-off gates, and local SQLite persistence.

What it fixes:Juggling 5 parallel agent terminal windows is chaos. You lose track of stuck tasks, background errors, and human approval prompts. Minions brings visual order.
Multi-agent parallel workflowsLong-running tasksHuman-in-the-loop approvalsLocal-first SQLite logging
npx minionsai
GitHub Repository
#12
572
Internet & Research

youtube-skills

ZeroPointRepo/youtube-skills

YouTube video transcript extractor, channel scanner, and playlist analyzer that bypasses cloud IP bans without yt-dlp blocks.

What it fixes:Agents repeatedly fail when asking for YouTube video summaries because cloud providers block raw scrapers. This skill provides reliable transcript ingestion.
Video tutorial summariesConference talk notesCompetitor video research
hermes skills install skills-sh/ZeroPointRepo/youtube-skills/skills/youtube-full
GitHub Repository
#13
291
Orchestration & Control

oh-my-hermes

witt3rd/oh-my-hermes

Multi-agent orchestration framework built natively for Hermes primitives: enforces research -> interview -> ralplan -> ralph execution cycles.

What it fixes:Prevents single-agent freestyling on complex architecture tasks where agents declare victory without validating requirements or edge cases.
Unfamiliar technical domainsComplex system designMulti-day feature rollouts
hermes skills install skills-sh/witt3rd/oh-my-hermes
GitHub Repository
#14
Skip Initially65
Media & Specialists

Resemble AI Detect

resemble-ai/detect-skill

Media safety verification skill detecting AI-generated or manipulated audio, synthetic voices, and deepfake video clips using Resemble's enterprise detection model.

What it fixes:Prevents agents and automation pipelines from citing or reposting fabricated media or deepfakes in research and content pipelines.
Content moderationBrand safetyInvestigative verification
npx skills add resemble-ai/detect-skill
GitHub Repository
#15
Unverified
Orchestration & Control

Composio skills

composiohq/skills

Tool router, auth management, and production session patterns connecting agents safely to 200+ SaaS platforms (Slack, Gmail, GitHub, Hubspot).

What it fixes:Writing raw OAuth and API integration code inside agent sessions is messy and fragile. Composio standardizes authentication and session handling.
CRM automationSlack/Email botsExternal app mutations
npx skills add composiohq/skills
#16
Unverified
Orchestration & Control

Browser Harness

Category Recommendation

Architecture pattern where an agent navigates your logged-in browser session and accumulates reusable site helpers for future runs.

What it fixes:Eliminates having to re-authenticate or re-teach navigation paths across websites every time an agent runs a browser task.
Authenticated site automationComplex internal portalsDynamic web scraping
Verify source before granting session permissions
#17
Unverified
Memory & Self-Improvement

Loopy / Loop Library

Category Recommendation

Loop discipline framework: enforces explicit success checks, hard stop conditions, and strict iteration caps for autonomous loops.

What it fixes:Unbounded agent loops can burn thousands of dollars in API credits and produce hundreds of hallucinated commits overnight.
Autonomous refactoringLong-running test-and-repair loopsCost governance
Define check rules, stop conditions & iteration caps
The Hidden Engineering Constraint

The Context Budget Problem Nobody Mentions

Skills consume context when they are available, not just when they execute. Descriptions, parameter schemas, and system triggers all occupy the prompt token window before your task even begins.

817 Skills

Cybersecurity Library

Overwhelms prompt memory if loaded on general frontend or copywriting tasks.

700+ Files

OpenMontage Pack

Consumes vast context budgets; only load inside dedicated video pipelines.

Zero Overhead

Matt Pocock & Day-One

Small composable skills loaded per repo keep agent attention sharp.

3 Golden Rules for Context Window Governance:
  1. Install per project, not globally: A video project does not need red-team security playbooks.
  2. Prune what never fires: If a pack hasn't triggered in 30 days, it is paying rent with your scarce prompt tokens.
  3. Favor composable packs: Prefer smaller, editable skill modules over giant monolithic frameworks.

Omitted Foundational Giants

The two largest skill repos on GitHub were excluded from this list because they are foundational base frameworks rather than external add-on skills:

obra/superpowers

Core foundational agent harness

★ 278,274

anthropics/skills

Official Anthropic reference skills

★ 171,918

Notable External Mentions

High-value specialized skill packages worth exploring for specific workflows:

Leonxlnx/taste-skillAnti-slop UI design★ 81,181
mvanhorn/last30daysRecent social synthesis★ 59,370
coreyhaines31/marketingskillsCRO & copywriting★ 45,797
kepano/obsidian-skillsObsidian vault bridge★ 47,352
Decision Protocol

How to Choose in 60 Seconds

Failure-driven installation beats aspirational installation every single time.

01. By Frequent Failure

What breaks most often?

Install only the skill for that exact failure point and test it for one week. Stop installing speculative toolchains.

02. By Workflow Type

Repetitive or Exploratory?

Repetitive work needs loops & mission control (Minions, codebase-memory). Exploratory work needs Agent-Reach & Defuddle.

03. By Public Exposure

Is output public-facing?

If code or copy is published publicly, Humanizer is mandatory to strip synthetic formatting and match your genuine voice.

Reference Architecture

Glossary of Agent Terms

Agent Skill

A written procedure an agent loads on demand when a task matches it. Adds consistency, not raw capability.

Hermes Agent

The Nous Research agent runtime these skills target, though most work across Claude Code, Codex, and Cursor too.

MCP (Model Context Protocol)

The standard connectivity protocol for exposing tools and structured data to AI agents.

Knowledge Graph

A structured map of entities and relationships used to understand codebases without token-heavy file reads.

Context Window

Working memory available for a task. Loaded skills consume context whether they execute or not.

Skill Pack

A packaged repository containing multiple related procedures installed as a modular suite.

Harness

The runtime wrapper handling agent tools, system prompts, memory, execution gates, and error recovery.

Agent Loop

A cycle where an agent plans, executes, checks results, and repeats until a defined stop rule is satisfied.

AI Tells

Predictable phrasing patterns marking text as machine-written (hype adjectives, rigid parallelism, vague abstractions).

Dual-Use Tooling

Techniques usable for defense or attack (security playbooks) requiring strict operational authorization.

Common Questions

Frequently Asked Questions

Detailed answers on context window impact, runtime compatibility, licensing, and installation.

Implementation Audit

Need Custom Multi-Agent Architecture?

I architect bespoke autonomous multi-agent pipelines, custom skill suites, and self-healing automation infrastructure for growing engineering teams.

Direct Calendar ScopingInstant Booking
Claim Free 15-Minute Scoping Session
or drop details below

You will speak directly with Muhammad Shadab Shams. Best fit: teams seeking automated workflows, custom internal operations tools, or AI integration. Get a free custom automation flowchart of your current workflow during our call.

No spam. Scoping response within 24 hours.