Hermes Agent includes a skill system. A skill is a markdown file that teaches the agent how to perform a task step by step. The author, Dibby, states that he builds apps daily with Hermes Agent and has generated $47,000 in less than 50 days using these AI tools. After reviewing hundreds of skills, he considers most of them low value. He presents his top 14.

The author gives two selection criteria. First, a well-built Hermes skill stays under 15 kilobytes. Larger skills increase agent memory usage and consume additional tokens.

Skills are installed by copying the GitHub repository and asking the Hermes agent to install it. An installed skill stays in the agent's memory. It can trigger automatically when a prompt matches its task, or the user can invoke it directly by instructing the agent to use it.

The list is ordered from 14 to 1. It falls into three groups: efficiency tools, capability expanders, and infrastructure tools.

Efficiency Tools

14. Skill Claw

• Popularity: More than 2,000 GitHub stars.
• Function: After each session, Skill Claw prompts the agent to run an evolution loop. The loop reviews the skills the agent used, merges overlapping skills, rewrites weak ones, and updates the skill library automatically.
• Effect: The agent improves over time without manual maintenance. The author states that the improvement compounds. In week one, the changes are minor. By week four, the agent differs substantially from its original setup.

13. Matt Pocock's Skills Pack

Matt Pocock, who runs a newsletter with more than 60,000 readers, released a pack of 15 skills. The author highlights three.

• Grill Me: Before the agent writes code, it asks five targeted questions about the intended build. The skill addresses a common cause of incorrect builds: insufficient context provided to the agent.
• Caveat: The skill removes token bloat from long sessions and reduces token usage by up to 75%. It is free.
• Teach Me: The agent teaches a subject and organizes practical takeaways in HTML. Pocock reports using it to learn how to solve a Rubik's Cube. The author notes that Pocock has a teaching background.

12. Defuddle

• Function: Defuddle converts web pages into clean, reader-mode markdown before the agent processes them. It removes navigation menus, footers, cookie banners, and sidebar advertisements.
• Rationale: Unprocessed pages consume tokens without adding useful content. Markdown is the format agents use to structure information.
• Effect: The author states that the agent reads the web three to four times more efficiently.
• Use cases: Research, competitive analysis, market research, and documentation lookups.

11. Humanizer

• Source: The skill is built into Hermes. The author reports that few users employ it.
• Problem: Agent-generated copy, such as emails, proposals, and social media posts, often contains recognizable AI patterns. Readers identify these patterns quickly. The author states that he stops reading when he sees an em dash.
• Function: Wikipedia maintains a page titled "Signs of AI writing," which documents common markers of AI-generated text. A developer converted that page into a skill that updates automatically when the page changes. The Humanizer skill rewrites agent output in a natural human voice.
• Use case: The author recommends it as a final review step for AI-assisted content before publication.

Capability Expanders

10. YouTube Full

• Problem: The default YouTube skill in Hermes fails on VPS and cloud environments because YouTube blocks cloud IP addresses.
• Function: YouTube Full replaces the default skill. It supports transcript extraction, channel browsing, playlist parsing, and video search. It does not require a Google API key.
• Infrastructure: The skill uses an API that processes 15 million transcripts per month.
• Use case: The author uses YouTube transcripts to train agents on writing YouTube scripts.

9. Composio

• Popularity: More than 20,000 GitHub stars.
• Problem: Connecting an agent to services such as Gmail requires complex setup, including Google Cloud Console configuration and OAuth. The effort multiplies with each additional tool.
• Function: Composio connects the agent to more than 1,000 SaaS tools, including Gmail, Google Sheets, Slack, Notion, HubSpot, and Salesforce. It removes the need for manual OAuth implementation and API key management.
• Effect: One skill allows the agent to operate the full business software stack. The author states that integrations were previously the main bottleneck.

8. Agent Skills by Addy Osmani

• Popularity: About 65,000 GitHub stars, the highest of any skill on the list.
• Author: Addy Osmani, senior engineering manager at Google Chrome.
• Content: A pack of 24 production-grade skills, accessed through eight slash commands that cover the development lifecycle:
  • /spec: Define requirements before writing code.
  • /plan: Break the work into tasks.
  • /build: Implement the code.
  • /test: Verify the implementation.
  • /review: Review the work before merging.
  • /ship: Deploy.
• Doubt-driven development: During a task, the agent pauses at each major decision. It lists its assumptions, challenges each one, and resolves gaps before proceeding. For production code or irreversible actions, the agent evaluates its approach before executing it. The author states that most agents skip this step unless instructed to perform it. The skill makes it automatic.

7. Resemble AI Detect

• Problem: Agents ingest user-submitted and web content. A significant share of this content may be AI-generated. Some content is designed to manipulate agent output.
• Function: The skill adds deepfake detection to ingestion pipelines. It detects AI-generated audio, images, video, and text. It identifies the generating tool, such as ElevenLabs, ChatGPT, or Claude. It labels detected content as AI-generated for the agent.
• Use case: Agents that scrape the web or search for specific information.

Infrastructure Tools

6. Mission Control

• Popularity: About 3,700 GitHub stars.
• Problem: Users running multiple Hermes agents, for example for research, coding, and client communication, lack visibility into agent activity. They cannot track costs or dispatch tasks across agents without managing each terminal separately.
• Function: Mission Control provides a fleet dashboard with task dispatch, agent health monitoring, real-time cost tracking, and live status across all agents.

5. Open Montage

• Popularity: About 12,000 GitHub stars.
• Description: The author describes it as the first open-source agentic video production system. It includes 12 pipelines, 52 tools, and more than 500 agent skills in a single installation.
• Workflow:
  1. The user provides a reference video, such as a YouTube video, short, or reel.
  2. The agent analyzes the transcript, pacing, scene structure, key frames, and tone.
  3. Before generating any frames, the agent presents two to three differentiated concepts, the full planned tool path, a cost estimate, and a preview.
  4. The user approves a direction before generation begins.
• Example: A 60-second Pixar-style animated short requires six generated motion clips, narration, music, and word-level captions. The estimated cost is $1.33.

4. Anthropic Cybersecurity Skills

• Author: Security researcher McColl 975.
• Content: More than 700 structured skills mapped to the MITRE ATT&CK framework, a knowledge base of adversary tactics and techniques based on real-world observations. The skills cover threat modeling, vulnerability assessment, secure code review, and incident response playbooks. The agent can query all of them.
• Rationale: Solo developers and small teams typically cannot hire a dedicated security engineer. The author states that teams handling real user data still need baseline security coverage. He presents this skill pack as a lower-cost substitute that covers the basics.

3. O My Hermes

• Inspiration: The O My Claude skill, which has 36,000 GitHub stars.
• Problem: AI agents often agree with user statements without verifying output. They accept a claimed problem or a claimed fix without evidence.
• Function: The skill converts a single agent session into a coordinated multi-agent workflow.
  • It breaks tasks into subtasks.
  • It assigns specialist agents or external CLI workers, including Codex, Gemini, and Cursor.
  • It runs tasks in parallel or in staged pipelines.
  • It verifies results before returning an answer.
• Stated benefits:
  • Faster execution of large tasks.
  • Less manual prompting.
  • Review and quality assurance through specialist roles.
  • Live visibility through session logs.
  • Token savings of up to 50% through smart routing.
  • Persistent execution until verification passes.
• Use cases: Shipping features, writing technical proposals, and planning migrations.

2. Details That Make Interfaces Feel Better

• Source: An article that has been shared thousands of times. A developer converted its recommendations into a skill.
• Function: Once installed, the agent applies the following techniques when building user interfaces:
  • Text wrapping that prevents headlines from leaving a single word on the last line.
  • Concentric border radius so that nested elements, such as a button inside a card, have matching corners.
  • Contextual icon animations using opacity, scale, and blur on interaction.
  • Tabular numbers so that changing values do not shift layout.
  • Interruptible animations that do not freeze when the user clicks quickly.
• Rationale: Agents do not apply these techniques unless instructed to do so.

1. Agent Reach

• Popularity: About 38,000 GitHub stars.
• Problem: Agents reason and build effectively but cannot access many internet sources. Twitter requires a paid API that most agents cannot use. Reddit returns HTTP 403 errors to cloud IP addresses. YouTube blocks requests from VPS environments.
• Function: Agent Reach provides access to Twitter, Reddit, YouTube, and GitHub with no API fees.
• Resilience: When a platform blocks one integration, Agent Reach uses a pre-mapped backup path. The agent continues operating without interruption.
• Effect: The author states that the other skills improve tasks the agent can already perform. Agent Reach adds a capability the agent lacked: access to a broader portion of the internet.
• Installation: Instruct the agent to install Agent Reach and provide the repository link.

Honorable Mentions

• Browser Harness: About 15,000 GitHub stars. Developed by the browser views team. It connects the agent to the user's Chrome browser, allowing the agent to click, scroll, fill forms, and navigate pages. When the agent encounters an unfamiliar page, it writes the missing helper function and continues.
• Codebase Memory MCP: About 11,800 GitHub stars, developed by a startup named Juice. It indexes an entire codebase into a persistent knowledge graph.
  • Indexing an average repository takes milliseconds.
  • In testing on the Linux kernel, 28 million lines of code were scanned in 3 minutes.
  • Agents use 120 times fewer tokens when querying the graph instead of reading files one by one.
  • The tool supports up to 158 programming languages.
• Loop Library: Developed by Matt Berman. Standard prompts request a single action. A loop defines how to evaluate results and repeat the action. The loop measures the outcome, keeps changes that improve results, and repeats until a target is met or improvement stops. For example, instead of "make this website faster," a loop instructs the agent to find the slowest page, make one focused improvement, measure the result, keep the change only if it helps, and repeat until every page meets the target. The skill provides a catalog of pre-built loops. If no loop fits, the agent helps the user design one in plain language.