The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is built around executing shell commands to manage tmux panes and git worktrees. It specifically utilizes a plan.json configuration that includes a launcherCommand field, which serves as a template for shell command execution (e.g., codex exec --cwd {worktree_path}).- [EXTERNAL_DOWNLOADS]: The instructions direct the user to install the dmux orchestration tool globally via npm (npm install -g dmux) from an external GitHub repository (github.com/standardagents/dmux).- [INDIRECT_PROMPT_INJECTION]: The orchestration patterns create a surface for cross-agent instruction injection through shared files and handoff mechanisms.
Ingestion points: The agent context is populated by plan.json, worker handoff.md files, and shared research outputs like /tmp/rate-limit-research.md.
Boundary markers: There are no explicit instructions for using delimiters or 'ignore embedded instructions' warnings when one agent consumes the output of another.
Capability inventory: The skill possesses high-privilege capabilities including spawning shell processes via tmux and performing complex git operations.
Sanitization: There is no evidence of sanitization or validation for the content generated by workers before it is ingested by subsequent agents in the workflow.

dmux-workflows