Agent Publish Workflow

A frictionless utility skill designed for "Vibe Coders" to securely automate the release preparation of AI Agent Skills. It sanitizes configuration data project-wide, performs a pre-flight secret scan to prevent data leaks, and packages the clean output without destroying the developer's local master files.

Core Features

• Non-Destructive Packaging: Choose your publish mode (dist folder, .zip archive, or direct overwrite). The workflow safely copies, sanitizes, and packages the project without overwriting your local settings (e.g. your private LLM endpoints or Anthropic API Keys).
• Pre-Flight Secret Scanner: Automatically hunts for exposed API Tokens (OpenAI, Anthropic, GitHub) and hardcoded passwords within your .md, .json, .py, and .txt files.
• Audit Logging: Every scan generates a timestamped .log report in the reports/ directory providing a paper trail of the pre-flight checks.
• Agentic Configuration Wiper: Dynamically detects <!-- CONFIGURATION_START --> blocks in all documentation files and resets them to the open-source factory default (UNCONFIGURED / N/A).
• Heavy Project Recognition: Built-in safeguards abort recursive scans if large software environments (like node_modules, .sln, or package.json) are accidentally targeted.

Usage

When triggering this skill via your agent (Cursor, Claude Code, etc.), it will:

1. Verify if your target project is an Agent Skill or a heavy Software Project.
2. Run the Pre-Flight Secret Scan against all readable text files (< 1MB).
3. Alert you if potential secrets are found.
4. Read your preferred Publish Mode.
5. Wipe the cloned configurations and finalize the export artifact (e.g., releases/release_vX.zip).

⚠️ Security Limitations

This is a pragmatic guardrail, NOT an Enterprise Data Loss Prevention (DLP) solution. The scanner relies on Regex to catch common slip-ups (e.g. pasted OpenAI keys). It does NOT recognize custom internal IPs, proprietary passwords, or complex business logic. You must still manually verify your files before pushing.

🤖 Agent Compatibility & Platform Notes

• Operating System: Currently, the logic is 100% PowerShell-based. It runs natively on Windows. Mac/Linux users must have PowerShell Core installed, or wait for the upcoming bash (.sh) equivalents.
• Directives: You will notice // turbo-all or // turbo comments above the script blocks in the .md files. This is a specific directive for the Antigravity Agent to auto-run the commands without asking for permission. If you use Cursor or Claude Code, these agents will simply ignore the comment and run the code normally.

Files

• SKILL.md - English instructions and trigger configuration.
• SKILL.de.md - German instructions and trigger configuration.

Installation

Clone this repository into your agent's skill directory (e.g., ~/.gemini/antigravity/skills/agent-publish-workflow) and reference it in your assistant to begin safely shipping your Agent architectures.