Publish Skill

Prepare, safety-review, version, commit, and publish local Codex skills through a GitHub-backed release flow and ClawHub CLI. Use when the user asks to host,...

Overview

Use this skill to move a local Codex skill from a working folder to a published, versioned artifact. Keep GitHub as the source of truth, run a local safety review before publication, and use ClawHub only after authentication and command details are clear.

Local Defaults

• Default GitHub account: jichengkai.
• Default GitHub profile: https://github.com/jichengkai.
• Prefer creating and pushing skill repositories under jichengkai unless the user explicitly names a different owner or organization.
• Treat this as a local identity preference only. It does not prove authentication and does not replace gh auth login, Git credentials, clawhub login, or CLAWHUB_TOKEN.

Boundaries

• Never ask for or expose GitHub passwords, ClawHub tokens, API keys, SSH private keys, or recovery codes in chat.
• If authentication is missing, ask the user to complete gh auth login, clawhub login, or secret setup locally; do not invent credentials.
• Treat publishing commands as state-changing. Explain what will be published before running them, and stop if the target path is ambiguous.
• Verify current official ClawHub documentation before relying on exact CLI flags when network access is available or the user asks for a real publish.
• Do not publish a skill that reads sensitive directories, handles credentials, runs obfuscated code, or makes undisclosed network calls unless the user explicitly accepts the risk after review.

Workflow

1. Locate the target skill.

   • Prefer an explicit path from the user.
   • Otherwise check the current workspace, then ${CODEX_HOME:-$HOME/.codex}/skills/<skill-name>.
   • Read the target SKILL.md, agents/openai.yaml if present, and a concise file listing before making decisions.

2. Validate the skill structure.

   • Use the local skill validator when available: /Users/jichengkai/.codex/skills/.system/skill-creator/scripts/quick_validate.py <skill-path>.
   • Confirm the skill has only needed files: SKILL.md, optional agents/, and purposeful scripts/, references/, or assets/.
   • Do not add README.md, CHANGELOG.md, or other auxiliary docs inside the skill unless the publishing platform explicitly requires them.

3. Run a pre-publish safety review.

   • Run python3 <this-skill>/scripts/review_skill.py <skill-path>.
   • Read references/security-review.md when the scanner reports findings, the skill contains executable code, or the user asks for a security pass.
   • Inspect scripts manually for destructive commands, credential access, hidden network behavior, and obfuscation.

4. Prepare the GitHub source repo.

   • Read references/publishing.md for the detailed GitHub and ClawHub flow.
   • Use jichengkai as the default GitHub owner for local publishing unless the user overrides it.
   • Check repository status before editing or committing, and preserve unrelated user changes.
   • Keep the repository history as the durable version record. If version metadata exists, bump it intentionally; otherwise use clear commit messages and release notes.

5. Publish or update.

   • Use clawhub skill publish <path> --slug <slug> --name "<display name>" for a direct publish after auth is present; add --owner <handle> for an org owner.
   • For repeatable automation, add a GitHub Actions workflow only when the user asks, and use a GitHub secret named CLAWHUB_TOKEN.
   • After publish, report the repository, commit, command run, result, and any manual auth step the user still needs.

Resource Guide

• references/publishing.md: detailed commands, versioning choices, and GitHub Actions pattern.
• references/security-review.md: manual safety checklist and blocker definitions.
• scripts/review_skill.py: local scanner for secrets, sensitive paths, network behavior, obfuscation hints, and unexpected binaries.

Final Response Checklist

• Target skill path and repo path.
• Validation and safety-review result.
• Git commit or branch created, if any.
• Publish command run or exact manual command still needed.
• Remaining auth or secret setup, without exposing any secret values.