ClawHub Skill Publisher

ClawHub Skill Publisher v1

> Turn a rough skill idea into a polished, publish-ready ClawHub listing — informed by what's actually working in the marketplace.

Use this skill when you want to:

Workflow

Step 1 — Research Top Listings

Install the most relevant published skills in a temp directory and read their SKILL.md + README.md:

mkdir -p /tmp/ch-research
Search for skills in your category
clawhub search "your-category-keyword"
Install top 3-5 results for analysis
clawhub install  --dir /tmp/ch-research --force
clawhub install  --dir /tmp/ch-research --force
(rate limit: add 3s sleep between installs)

What to capture per skill:

Step 2 — Gap Analysis

Compare your draft against findings. Score each dimension:

| Dimension | Best Practice | Your Draft | Action | |-----------|---------------|------------|--------| | Description line | ≤160 chars, value-first, no buzzwords | ? | Patch or OK | | "When to use" | Explicit trigger + do/don't | ? | Patch or OK | | Commands/interface | Slash commands or trigger phrases | ? | Patch or OK | | Word count (SKILL.md) | 400–700 words | ? | Trim or expand | | Tables vs. prose | Tables preferred for comparisons | ? | Patch or OK | | Version history | Present, at bottom | ? | Add or OK | | Safety section | Explicit "never does X" list | ? | Add or OK | | Examples | Concrete ✅/❌ pairs | ? | Add or OK | | Attribution | Link back to openclaw.ai / clawhub.ai | ? | Add or OK |

Step 3 — Patch the Draft

Apply gap findings. Priority order: 1. Description line (most visible — fix first) 2. "When to use" section (drives installs) 3. Trim word count if over 700 (cut prose, keep tables) 4. Add missing sections (safety, version history) 5. Convert prose comparisons to tables 6. Add examples file if none exists

Step 4 — Publish

# Verify auth
clawhub whoami
Publish (run from workspace root or skill parent dir)
clawhub publish ./skills/ \
  --slug  \
  --name "Your Skill Name" \
  --version 1.0.0 \
  --changelog "Initial release"

Published URL: https://clawhub.ai/skills/

ClawHub Listing Anatomy

Description Field (≤160 chars)

Formula: [What it does] + [how] + [key outcome].

✅ Good: "Reduce AI costs by batching related asks into fewer responses. ~30–50% fewer API calls, no quality loss." ❌ Bad: "ClawSaver — Combines Linked Asks into Well-structured Sets for Affordable, Verified, Efficient Responses"

SKILL.md Structure (what the agent reads)

---
name: skill-name
version: X.Y.Z
description: "Same as listing description"
metadata: {"openclaw":{"emoji":"🔧"}}
Skill Name vX
> One-line positioning statement.
[One paragraph: what it does and why.]
When to Use
[Use / Do not use — explicit conditions]
Core Behavior / Commands
[Tables preferred. Trigger phrases, commands, decision rules.]
Safety
[What it never does. Explicit opt-outs.]
Installation
[clawhub install command]
Version History
[- X.Y.Z — what changed]

README.md Structure (humans + listing body)

# Skill Name
> Tagline
Why [Skill Name]?
[Problem → solution in 2-3 sentences]
What It Does
[Numbered or bulleted feature list]
[Key Decision Table or Usage Example]
Safety Model
Installation
Version

Marketplace Patterns (Observed Feb 2026)

What top skills have in common

What separates good from great

Category density (as of Feb 2026)

File Checklist Before Publishing

Skill Type: Behavior-Change vs. Active Tool

Most ClawHub skills are behavior-change skills — they work by shaping agent judgment through instructions, not by running code or intercepting requests at the system level. This is the same mechanism as execution-loop-breaker, token-saver, and most top listings.

When writing a behavior-change skill:

When a skill needs to be an active tool:

Bottom line: Instruction-based skills are legitimate and valuable. Just be honest about the scope. Users trust skills that set accurate expectations.

Version History

[Use / Do not use — explicit conditions]