CLAWHUB PUBLISH WORKFLOW 📦

Complete workflow for publishing skills to ClawHub

Info	Value
Version	1.3.0 — 2026-05-07
Status	OPERATIONAL

1. PURPOSE AND SCOPE

Objective

Provide a complete workflow for any OpenClaw agent to publish skills to ClawHub using dual evaluation systems.

Universal Design

Works for ANY OpenClaw agent, regardless of cluster or environment.

When to Use

Trigger	Action
Creating a new skill to publish	Follow this workflow
Need to evaluate and improve	Use bundled dual evaluator
Publishing to ClawHub	Follow all steps
Fixing a failed evaluation	Run evaluation + fix

2. BUNDLED TOOLS

File	System	Purpose
`evaluator.py`	5-Dimension	Colorful 5-dimension evaluation
`eval-skill.py`	ISO 25010	Automated structural checks (13 tests)
`references/tools.md`	—	Complete tools reference

Dual Evaluation System

┌─────────────────────────────────────────────┐
│         CLAWHUB PUBLISH WORKFLOW            │
├─────────────────────────────────────────────┤
│  1. eval-skill.py (ISO 25010)              │
│     → Automated checks (13 tests)          │
│     → Target: 90%+ structural score         │
│                                             │
│  2. evaluator.py (5-Dimension)              │
│     → 5 dimensions × 20% = 100              │
│     → Target: 70+ score                     │
│                                             │
│  3. Manual (25 criteria)                   │
│     → ISO 25010 rubric (8 categories)       │
│     → Target: 80+ score                     │
└─────────────────────────────────────────────┘

3. COMPLETE WORKFLOW

Step 1: Create SKILL.md

TASKS:
1. Create SKILL.md in English (impersonal, universal)
2. Include frontmatter with name + description
3. Follow standard skill structure
4. Make it addressable to ANY agent

Frontmatter Template:

---
name: skill-name
description: "Skill description. Use when: (1) trigger1, (2) trigger2, (3) trigger3. Provides X, Y, Z."
---

Step 2: Automated ISO 25010 Checks

# Run automated structural checks
python3 eval-skill.py <skill-path> --verbose

# Target: 90%+ (12+/13 checks passed)

Step 3: 5-Dimension Evaluation

# Run full 5-dimension evaluation
python3 evaluator.py <skill-path> --verbose --improve

# Target: 70+ score

Step 4: Manual 25-Criteria Assessment

Score using the rubric:

Category	Max	Score
Functional Suitability	/12
Reliability	/12
Performance	/8
Usability (AI)	/12
Usability (Human)	/8
Security	/12
Maintainability	/12
Agent-Specific	/24
TOTAL	/100

Step 5: Fix Issues

Low Score Area	Fix
Structure <15	Add missing sections, formatting
Clarity <15	Add examples, commands, constraints
Completeness <15	Add tools, prerequisites, errors
Consistency <15	Add style markers, naming consistency
Functionality <15	Fix command syntax

Step 6: Re-evaluate

# Continue until targets met
python3 eval-skill.py <path> --verbose  # Target: 90%+
python3 evaluator.py <path> --verbose    # Target: 70+

Step 7: Publish

# Login
clawhub login --token <token> --no-browser

# Check slug availability
clawhub inspect <slug> 2>&1

# Publish
clawhub publish <path> \
  --slug <slug> \
  --name "<Name>" \
  --version 1.0.0 \
  --changelog "<changes>"

Step 8: Verify

clawhub inspect <slug>

4. EVALUATION TARGETS

System	Metric	Target	Command
ISO 25010	Structural	90%+	`eval-skill.py --verbose`
5-Dimension	Score	70+	`evaluator.py --verbose`
Manual	25-Criteria	80+	Rubric scoring

5. COMMON ISSUES

"Slug already taken"

→ Use alternative: yourname-<skill-name>

"API token invalid"

→ Get new at https://clawhub.com/settings/tokens

"Name mismatch"

→ Rename directory to match frontmatter name:

Score < 70 after fixes

→ Focus on lowest scoring dimensions first

6. EDGE CASES

Case	Solution
No frontmatter	Add `---` at start with name + description
Missing tools section	Document all required tools
No examples	Add at least 2 concrete examples
Very long SKILL.md	Split into references/ if needed
Empty scripts dir	Remove empty directories

7. SKILL STRUCTURE TEMPLATE

---
name: skill-name
description: "Skill description. Use when: (1) trigger1, (2) trigger2."
---

# SKILL NAME

## 1. PURPOSE AND SCOPE

## 2. BUNDLED TOOLS

## 3. PREREQUISITES

## 4. USAGE
```bash
# Commands here

5. EXAMPLES

6. CONSTRAINTS

7. ERROR HANDLING

8. EDGE CASES


---

_In Altum Per Partage._
📦 ClawHub Publish Workflow v1.3

clawhub-publish-workflow

Safety Notice

Copy this and send it to your AI assistant to learn