The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines an orchestration workflow where a script (scripts/orchestrate-worktrees.js) processes a plan.json file containing worker instructions.
Ingestion points: Data is ingested from plan.json (SKILL.md) to define worker tasks and execution commands.
Boundary markers: The documentation does not specify the use of delimiters or 'ignore' instructions for the content of the plan.json fields.
Capability inventory: The orchestration script can execute shell commands (launcherCommand), manage git worktrees, and write to the local filesystem.
Sanitization: There is no evidence of input validation or sanitization for the commands or task strings provided in the configuration file.
[COMMAND_EXECUTION]: The workflow relies on executing shell commands provided within a JSON configuration. The launcherCommand property allows for the execution of arbitrary tools (e.g., codex exec, bash), which represents a command injection risk if the JSON source is untrusted.
[EXTERNAL_DOWNLOADS]: The documentation directs users to install the dmux tool from a third-party GitHub repository (github.com/standardagents/dmux).
[DYNAMIC_EXECUTION]: The skill describes a process where a script dynamically generates file content (task.md, handoff.md) and launches worker processes based on a provided plan template.

dmux-workflows