dmux-workflows

Pass

Audited by Gen Agent Trust Hub on Oct 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The ECC helper functionality reads a plan.json file to orchestrate agent sessions. This represents a potential surface for indirect prompt injection if the configuration file is created or influenced by untrusted external data.
  • Ingestion points: The plan.json configuration file, particularly the launcherCommand and workers[].task fields in SKILL.md.
  • Boundary markers: No explicit delimiters or safety instructions are mentioned to isolate external task data from the command templates.
  • Capability inventory: Execution of shell commands through the orchestrate-worktrees.js script.
  • Sanitization: The instructions do not specify any validation or sanitization routines for the content of the plan.json file.
  • [DYNAMIC_EXECUTION]: The scripts/orchestrate-worktrees.js script performs runtime command assembly. It uses a template provided in the launcherCommand field and interpolates dynamic values such as {worktree_path} and {task_file} before executing the resulting string in a shell pane.
  • [COMMAND_EXECUTION]: The skill relies on and instructs the user to execute various shell commands, including dmux, git worktree, and the orchestrate-worktrees.js script which spawns multiple agent processes.
  • [EXTERNAL_DOWNLOADS]: The skill references an external orchestration tool, dmux, and provides a link to its repository (github.com/standardagents/dmux) for installation. This introduces an external dependency from a third-party source.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 14, 2026, 03:13 PM