The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The documentation recommends downloading and installing the dmux tool from a third-party GitHub repository (github.com/standardagents/dmux).
[COMMAND_EXECUTION]: The skill utilizes a local script scripts/orchestrate-worktrees.js that executes shell commands within tmux panes. These commands are derived from a plan.json file, allowing for arbitrary command execution based on the configuration.
[INDIRECT_PROMPT_INJECTION]: The orchestration workflow involves reading task instructions from a configuration file (plan.json) and passing them to sub-agents. This creates a surface for indirect prompt injection if the plan is generated from untrusted external data.
Ingestion points: The workers array and launcherCommand field in plan.json serve as data entry points.
Boundary markers: None identified in the provided documentation or examples.
Capability inventory: The system can spawn tmux sessions, execute shell commands via the launcher, manage git worktrees, and perform file writes for task management.
Sanitization: No evidence of input validation or sanitization for the task strings or command templates was found.
[DYNAMIC_EXECUTION]: The helper script uses template-based string interpolation (e.g., {worktree_path}, {task_file}) to dynamically construct and execute commands at runtime based on the orchestration plan.

dmux-workflows