The Agent Skills Directory

[COMMAND_EXECUTION]: The repo.py script manages the project lifecycle by executing shell commands for git and the GitHub CLI (gh). It handles repository creation, commits, branching (try/keep/drop workflow), and secret scanning. Commands are executed using standard argument lists to prevent injection.
[EXTERNAL_DOWNLOADS]: The skill downloads media assets from GitHub Releases and external URLs via urllib.request.urlretrieve to restore project state.
[REMOTE_CODE_EXECUTION]: While the skill downloads remote files, it implements strict integrity checks by verifying the SHA-256 hash of every downloaded asset against a local lockfile (media.lock.jsonl) before the file is placed in the project directory.
[CREDENTIALS_UNSAFE]: The skill includes a proactive secret scanner in repo.py that checks for hardcoded API keys (e.g., Anthropic, OpenAI, GitHub, etc.) and private keys. It blocks commits containing secrets and guides the user to use .env files, which are automatically added to .gitignore.
[INDIRECT_PROMPT_INJECTION]: The skill processes various project files (briefs, scripts, metadata) that could contain untrusted content. The documentation explicitly warns users about the risk of agents reading malicious AGENTS.md files in shared directories and provides templates for safe project structure.
[PRIVILEGE_ESCALATION]: The skill automatically sets up Git hooks for local safety checks, which involves setting executable bits on the hook scripts. This is a standard project configuration step and is not used for unauthorized permission changes.

video-project-github