The Agent Skills Directory

[COMMAND_EXECUTION]: Establishes system persistence and scheduled background execution.
Evidence: Modifies the user's crontab and creates/loads macOS launchd plist files in ~/Library/LaunchAgents/ to run automation scripts periodically.
[COMMAND_EXECUTION]: Encourages the bypass of built-in security controls.
Evidence: Explicitly prompts the user to consider using the --dangerously-skip-permissions flag for unattended runs, which removes human-in-the-loop approval for sensitive operations.
[COMMAND_EXECUTION]: Performs dynamic script generation and runtime execution.
Evidence: Generates, verifies, and executes helper scripts within ~/.claude/scripts/ and creates project-specific Taskfile.yml configurations.
[EXTERNAL_DOWNLOADS]: Fetches external code from GitHub.
Evidence: Uses the gh repo clone command to download entire repositories from specified organizations to the local filesystem.
[PROMPT_INJECTION]: Contains instructions to override default agent behavior.
Evidence: Includes a directive to "automatically exit" after completing phases if run as a standalone request.
[COMMAND_EXECUTION]: Potential for indirect prompt injection via external data.
Ingestion points: Reads CLAUDE.md files from potentially untrusted cloned repositories.
Boundary markers: None identified.
Capability inventory: Can write files, modify cron/launchd, and execute arbitrary shell commands via generated scripts.
Sanitization: None identified for the content read from CLAUDE.md before processing logic.

org-orchestrate