| name | skill-publisher |
| description | Create, validate, security-scan, and publish skills to ClawHub. Use when asked to "make a skill", "publish a skill", "create a new skill", "scaffold a skill", "package this as a skill", or any request to build and distribute agent capabilities.
|
Skill Publisher
Build and publish skills to ClawHub in one flow.
Quick Start
1. Scaffold a new skill
bash {baseDir}/scripts/scaffold.sh <skill-name> [--dir <output-dir>]
Creates a new skill folder with SKILL.md template and scripts/ directory.
Default output: ./skills/<skill-name>
2. Fill in the skill
Edit the generated SKILL.md:
- Set
name and description in frontmatter (description is critical for triggering)
- Write clear instructions in the body
- Add scripts/ for executable code, references/ for docs, assets/ for templates
3. Validate
bash {baseDir}/scripts/validate.sh <skill-folder>
Checks:
- Required files exist (SKILL.md)
- Frontmatter has
name and description
- Naming conventions (lowercase, hyphens)
- No forbidden files (README.md, CHANGELOG.md, etc.)
4. Security scan
bash {baseDir}/scripts/security-scan.sh <skill-folder>
Scans for red flags:
- Remote code execution / eval patterns
- Data exfiltration (curl to unknown hosts)
- Environment variable harvesting
- Prompt injection in markdown files
- Suspicious file permissions
5. Publish
bash {baseDir}/scripts/publish.sh <skill-folder> --slug <name> --version <x.y.z>
Pushes to ClawHub. Requires clawhub login first.
One-liner (for simple skills)
bash {baseDir}/scripts/scaffold.sh my-skill && \
bash {baseDir}/scripts/validate.sh skills/my-skill && \
bash {baseDir}/scripts/security-scan.sh skills/my-skill && \
bash {baseDir}/scripts/publish.sh skills/my-skill --slug my-skill --version 1.0.0
Skill Anatomy Reminder
my-skill/
├── SKILL.md ← Required. Frontmatter (name, description) + instructions.
├── scripts/ ← Optional. Executable code (bash, python, etc.)
├── references/ ← Optional. Docs loaded on-demand into context.
└── assets/ ← Optional. Templates, images, files used in output.
Key principles:
- Be concise. Context window is shared real estate.
- Description in frontmatter is the trigger — make it comprehensive.
- Progressive disclosure: SKILL.md body only loads when triggered.
- Scripts > inline code for deterministic, repeated operations.
