Publish Skill

Prepare and publish an OpenClaw skill to ClawHub. This skill codifies the audit → generalize → publish workflow.

When To Use

• Publishing a new skill to ClawHub
• Updating an existing published skill
• When the user says "publish this skill", "prepare for publishing", "make a publishable copy"
• NOT for installing skills from ClawHub (that's npx clawhub@latest install)

Workflow

Step 1: Audit the Live Skill

Before creating any copy, audit the source skill for secrets and PII:

1. Read every file in the skill directory recursively
2. Check for these categories of sensitive content:

3. Report findings to the user before proceeding — do not silently modify

Step 2: Create Publishable Copy

Create a separate directory (never modify the live skill):

$CLAWHUB_DEFAULT_DIR/<skill-name>-skill/

Default base: ~/projects/skills (override via CLAWHUB_DEFAULT_DIR env var).

Directory structure:

<skill-name>-skill/
├── SKILL.md           # Manifest (generalized)
├── README.md          # User-facing docs
├── .gitignore         # Standard ignores
├── scripts/           # Script files (generalized)
├── references/        # Optional reference docs
└── ...                # Any other skill-specific files

Step 3: Generalize Content

For each file in the skill:

SKILL.md frontmatter:

• Add env: block declaring all extracted env vars with descriptions and required/optional
• Remove any personal identifiers from description

Scripts (Python, Shell, etc.):

• Replace hardcoded paths with os.environ.get("VAR", fallback) / env var reads
• Replace hardcoded timezones with env var (UTC fallback)
• Remove now()/utc_now() fallbacks that bypass source timestamps — raise errors instead
• Remove personal data (medication names become empty lists with edit instructions, etc.)
• Remove dead code and unused imports

Documentation (Markdown):

• Remove Discord IDs, channel names, user IDs
• Replace personal examples with generic ones
• Keep timezone/ID references only as example values (e.g. "e.g. America/Los_Angeles")
• Remove internal URLs/IPs

Shell wrappers:

• Use relative path resolution: SCRIPT="$(cd "$(dirname "$0")" && pwd)/tracker.py"
• Remove hardcoded absolute paths