Trigger: When user asks for "weekly report" or "status update"

Steps:

Learned preferences:

• User prefers percentages over absolute numbers
• Include "Action Items" section at the end
• Keep total length under 500 words

Simple. Readable. And it accumulates over time.

Your First Custom Skill (Step-by-Step)

You don't have to wait for the learning loop. You can create skills manually — and for repeatable tasks, you should.

Step 1: Identify the Repeating Pattern

Good candidates for skills:

• Any task you run more than twice a week
• Any task with a consistent input → output structure
• Any task where you've corrected Hermes' approach before

• One-off creative tasks
• Tasks that change every time
• Tasks requiring real-time judgment calls

Step 2: Write the Skill File

Trigger: "check competitor prices" or "price monitoring"

Required inputs: product_urls (list)

Steps:

Output format: Markdown table with columns: Product | Price | Change | Status

Step 3: Test and Iterate

Run the skill once manually. Watch what Hermes does with it. The first run often reveals:

• Steps that need more detail
• Edge cases the skill doesn't handle
• Formatting preferences to add

Where Hermes Skills Hit Their Limits

Here's where honesty matters. The skill loop is great for tasks that stay within Hermes' existing capabilities. But for browser automation against real websites, native skills hit a wall.

The Skill Knows What to Do — But Can't Execute

A perfectly written Hermes skill for scraping Amazon might include every correct step: navigate, wait for rendering, extract prices, format output. The problem isn't the instructions — it's the execution environment.

The skill is smart. The browser is dumb. Without anti-detection fingerprints, CAPTCHA solving, and residential proxies, the skill's instructions are worthless against protected sites.

The Learning Loop Can't Learn What It Can't See

Hermes' reflection cycle learns from successes and failures. But against anti-bot protection, every attempt fails the same way — there's no successful variation to learn from. The loop generates skill files that encode different approaches to the same blocked request.

It's like a chess player trying to improve by replaying only lost games where they never made a legal move. There's nothing to learn when the entire board is inaccessible.

Static Sites vs. Modern Websites

For the top row, Hermes skills are sufficient. For everything else — which is an increasing percentage of the web — you need a different kind of skill.

10 Ready-Made BrowserAct Skills That Fill the Gap

BrowserAct Skills are a fundamentally different approach to the same problem. Instead of the agent learning through trial and error, someone has already figured out the correct extraction path for a specific website — and encoded it.

Each Skill handles the full chain: stealth browser → anti-detection → CAPTCHA solving → site-specific extraction → structured data output.

Here are 10 that solve the most common requests Hermes users run into:

E-commerce & Product Data

1. Amazon Product Search API — Search products, extract titles, prices, ratings, ASINs. Handles Amazon's dynamic rendering and anti-bot protection automatically.

2. Amazon Reviews API — Pull review text, ratings, dates, verified purchase status. Structured data ready for sentiment analysis.

3. Amazon Competitor Analyzer — Competitive pricing intelligence across product categories. The kind of data that takes hours manually.

Local Business & Maps

4. Google Maps API — Extract business listings, ratings, addresses, phone numbers, operating hours. Clean structured output from Google's heavily protected maps interface.

5. Google Maps Reviews API — Pull individual reviews with text, ratings, and dates. Essential for reputation monitoring and local SEO.

Video & Content

6. YouTube Video API — Video metadata, view counts, engagement metrics, descriptions. Works regardless of YouTube's frequent UI changes.

7. YouTube Channel API — Channel statistics, subscriber counts, upload history. No official API key required.

8. YouTube Comments API — Extract comments at scale for sentiment analysis, audience research, or content ideation.

News & Search

9. Google News API — Real-time news monitoring by keyword, topic, or source. Structured headlines, snippets, publication dates.

10. Google Image API — Image search results with source URLs, dimensions, and metadata. Useful for visual content research and competitive analysis.

All 10 are available on ClawHub — install in one click, works with Hermes via API or MCP.

Hermes Skills + BrowserAct Skills: Using Both Together

This isn't an either/or choice. The best Hermes setups use both skill systems — each for what it's good at.

How They Complement Each Other

The Integration Pattern

Hermes handles the workflow logic, scheduling, and messaging. BrowserAct handles the browser-based data extraction. The native skills process and format the data. The BrowserAct Skills get the data in the first place.

Connecting BrowserAct to Hermes via MCP

One configuration line in your Hermes setup:

Once connected, every BrowserAct Skill becomes available to Hermes as a tool. The agent doesn't need to know how the Skill works internally — it just calls it and gets structured data back.

Skill Management Best Practices

Whether you're using native Hermes skills, BrowserAct Skills, or both — a few patterns keep things clean.

1. Name Skills by Outcome, Not Process