The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends installing 'dmux' from a non-whitelisted GitHub repository (github.com/standardagents/dmux), which entails downloading and running third-party code.
[COMMAND_EXECUTION]: The orchestration helper script (scripts/orchestrate-worktrees.js) performs significant shell operations, including managing git worktrees (git worktree add) and launching multiple parallel agent processes based on a JSON-defined plan.
[COMMAND_EXECUTION]: The execution logic uses a launcherCommand template from plan.json to spawn workers; this pattern of dynamic command construction with variable interpolation (e.g., {worktree_path}, {task_file}) presents a command injection surface if the input file is untrusted.
[PROMPT_INJECTION]: The skill processes task descriptions from an external plan.json and supplies them directly to parallel agent sessions, creating an indirect prompt injection surface.
Ingestion points: plan.json (worker task definitions).
Boundary markers: None specified in the documentation.
Capability inventory: Shell command execution, file system management, and process spawning across multiple panes.
Sanitization: No evidence of input validation or sanitization for task content or launcher commands.

dmux-workflows