Documentation

Everything you need to integrate OpenByt skills into your AI agent workflow.

Quick Start

Get up and running with OpenByt skills in under 2 minutes. Skills are portable automation modules that plug into any compatible AI agent.

1. Purchase and download

Browse the Skills marketplace, add a skill to your cart, and complete checkout. You’ll receive an instant download link for your .skill.md file.

2. Add to your agent

Move the downloaded file into your agent’s skills directory:

bash

mv ~/Downloads/github-pr-workflow.skill.md ~/agent/skills/

3. Load and run

Your agent automatically detects new skills. Or load manually:

bash

hermes skill load github-pr-workflow

💡

Skills persist across sessions. No need to reload after restart — your agent remembers them.

Installation

OpenByt skills work with any agent that supports the standard skill format.

Prerequisites

AI Agent — Hermes Agent v2.0+, OpenClaw v1.5+, or compatible
Python 3.10+ — for skills with Python scripts
Git — optional, for version-controlled skill management

Directory structure

text

~/agent/
├── skills/           # Drop .skill.md files here
│   ├── github-pr-workflow.skill.md
│   ├── api-integration-builder.skill.md
│   └── ...
├── config.yaml       # Agent configuration
└── scripts/          # Auto-created by skills

Your First Skill

Let’s create a simple skill from scratch to understand the format. We’ll build a skill that generates commit messages.

Create the file

bash

touch ~/agent/skills/commit-message.skill.md

Write the skill

yaml

---
name: commit-message
version: 1.0.0
description: Generate conventional commit messages
triggers:
  - "write commit message"
  - "generate commit"
tools: [terminal]
category: development
---

# Commit Message Generator

## Steps
1. Run `git diff --staged` to see changes
2. Analyze the diff content
3. Generate a conventional commit message
4. Format: type(scope): description

## Rules
- Use present tense ("add" not "added")
- Keep subject line under 72 characters
- Types: feat, fix, docs, style, refactor, test, chore

Test it

bash

hermes skill load commit-message
hermes skill validate ./commit-message.skill.md

Now when you say “write commit message” to your agent, it will follow the steps defined in your skill.

Skill Format

Every skill follows a standard structure — YAML frontmatter for metadata, markdown body for instructions:

yaml

---
name: github-pr-workflow
version: 1.2.0
description: Automate GitHub PR creation and review
triggers:
  - "create a PR"
  - "open pull request"
tools: [terminal, web]
---

# GitHub PR Workflow

## Steps
1. Check current branch and changes
2. Create feature branch if needed
3. Push and open PR via gh CLI

## Pitfalls
- Never push directly to main
- Check for uncommitted changes first

Frontmatter reference

Field	Type	Description
`name`	string	Unique identifier (lowercase, hyphens)
`version`	semver	Version for update tracking
`description`	string	One-line summary
`triggers`	array	Activation phrases
`tools`	array	Required toolsets
`category`	string	Grouping (devops, marketing…)

Configuration

Control how your agent loads and manages skills via config.yaml:

Basic configuration

yaml

skills:
  directory: ~/agent/skills
  auto_load: true
  max_loaded: 50
  update_check: daily

Configuration options

Option	Default	Description
`directory`	`~/agent/skills`	Where skill files are stored
`auto_load`	`true`	Load all skills on agent startup
`max_loaded`	`50`	Maximum concurrent skills in memory
`update_check`	`daily`	How often to check for updates (daily/weekly/never)

Per-skill overrides

Override default behavior for specific skills:

yaml

skill_overrides:
  github-pr-workflow:
    default_branch: main
    auto_push: false
  kubernetes-deployer:
    namespace: production
    dry_run: true

Environment variables

Skills can access environment variables for secrets and configuration:

bash

# Set API keys for skills that need them
export GITHUB_TOKEN=ghp_xxxxxxxxxxxx
export OPENAI_API_KEY=sk-xxxxxxxxxxxx

# Or add to your agent's .env file
echo "GITHUB_TOKEN=ghp_xxx" >> ~/agent/.env

🔒

Never hardcode secrets in skill files. Use environment variables or your agent’s secret management system.

Triggers

Triggers define when a skill activates. The agent matches user input against trigger phrases using fuzzy matching.

Trigger types

Type	Example	Description
Exact phrase	`"create a PR"`	Activates on exact or near-exact match
Keyword	`"deploy"`	Activates when keyword appears in context
Pattern	`"review PR #*"`	Wildcard matching for dynamic input
Auto	`"auto:file_change"`	Event-driven activation (file changes, cron)

Priority and conflicts

When multiple skills match the same input, the agent resolves conflicts by:

Specificity — longer, more specific triggers win
Recency — recently used skills get a slight boost
Explicit override — use priority: high in frontmatter

Testing triggers

bash

# Test which skill matches a phrase
hermes skill match "create a pull request"

# Output: github-pr-workflow (score: 0.94)

💡

Keep triggers specific and unique. Avoid generic phrases like “help” or “do this” that could match multiple skills.

Hermes Agent

Hermes Agent has native support for OpenByt skills. Here’s how to integrate:

Setup

yaml

# ~/.hermes/config.yaml
skills:
  directory: ~/agent/skills
  auto_load: true
  sources:
    - name: openbyt
      type: marketplace
      api_key: YOUR_OPENBYT_API_KEY

Auto-update from marketplace

Configure Hermes to automatically check for skill updates:

bash

# Enable auto-updates
hermes config set skills.update_check daily

# Manual update check
hermes skills update --source openbyt

Skill permissions

Hermes enforces tool access based on the tools field in skill frontmatter. A skill requesting [terminal, web] can only use those toolsets — it cannot access browser, file system, or other tools unless explicitly declared.

Toolset	Access	Use case
`terminal`	Shell commands	Git, builds, scripts
`web`	HTTP requests	APIs, scraping
`browser`	Browser automation	Web interaction
`file`	File read/write	Config, data files

OpenClaw

OpenClaw supports the standard skill format with minor configuration differences:

Installation

bash

# Add a skill to OpenClaw
openclaw skills add ./github-pr-workflow.skill.md

# List installed skills
openclaw skills list

# Remove a skill
openclaw skills remove github-pr-workflow

Configuration

json

{
  "skills": {
    "directory": "./skills",
    "autoLoad": true,
    "marketplace": {
      "provider": "openbyt",
      "apiKey": "YOUR_KEY"
    }
  }
}

Compatibility notes

OpenClaw uses JSON config instead of YAML — skill files themselves remain the same
The auto: trigger prefix is not supported in OpenClaw v1.x — use cron-based scheduling instead
Tool names map 1:1 between Hermes and OpenClaw

REST API

Access purchased skills programmatically:

bash

# List purchased skills
curl -H "Authorization: Bearer YOUR_KEY"   https://api.openbyt.com/v1/skills

# Download a skill
curl -H "Authorization: Bearer YOUR_KEY"   https://api.openbyt.com/v1/skills/github-pr-workflow/download   -o github-pr-workflow.skill.md

Endpoints

Method	Endpoint	Description
`GET`	`/v1/skills`	List purchased skills
`GET`	`/v1/skills/:id`	Skill metadata
`GET`	`/v1/skills/:id/download`	Download file
`GET`	`/v1/skills/:id/updates`	Check for updates

CLI Commands

Complete reference for skill management via the command line.

Skill lifecycle

bash

# List all installed skills
hermes skills list

# Load a specific skill
hermes skill load github-pr-workflow

# View skill details and metadata
hermes skill view github-pr-workflow

# Validate skill format before loading
hermes skill validate ./my-skill.skill.md

# Remove a skill
hermes skill remove github-pr-workflow

Marketplace commands

bash

# Search marketplace
hermes skills search "kubernetes deploy"

# Update all skills to latest version
hermes skills update

# Update a specific skill
hermes skills update github-pr-workflow

# Check for available updates
hermes skills outdated

Debugging

bash

# Test trigger matching
hermes skill match "deploy to production"

# Run skill in dry-run mode (no side effects)
hermes skill run github-pr-workflow --dry-run

# Show skill execution log
hermes skill log github-pr-workflow --last 5

Command flags

Flag	Commands	Description
`--verbose`	all	Show detailed output
`--dry-run`	run, update	Preview without executing
`--force`	remove, update	Skip confirmation prompts
`--json`	list, view	Output in JSON format
`--source`	update, search	Filter by marketplace source

Troubleshooting

Skill not loading

Verify the file has valid YAML frontmatter and is in the correct directory. Run the validator:

bash

hermes skill validate ./my-skill.skill.md

Trigger conflicts

If a trigger phrase matches multiple skills, the agent picks the best match by specificity. Make triggers unique and descriptive to avoid collisions.

Permission errors

Check the tools field in frontmatter — your agent must have those toolsets enabled in config.yaml.

🔧

Still stuck? Reach out at info@openbyt.com or visit the Help Center.

Docs

Documentation

Quick Start

1. Purchase and download

2. Add to your agent

3. Load and run

Installation

Prerequisites

Directory structure

Your First Skill

Create the file

Write the skill

Test it

Skill Format

Frontmatter reference

Configuration

Basic configuration

Configuration options

Per-skill overrides

Environment variables

Triggers

Trigger types

Priority and conflicts

Testing triggers

Hermes Agent

Setup

Auto-update from marketplace

Skill permissions

OpenClaw

Installation

Configuration

Compatibility notes

REST API

Endpoints

CLI Commands

Skill lifecycle

Marketplace commands

Debugging

Command flags

Troubleshooting

Skill not loading

Trigger conflicts

Permission errors