The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on shell scripts (e.g., spawn_agents.sh, collect_results.sh, merge_patches.sh) to manage parallel worker agents. These scripts perform file system operations, use git worktree to create isolated project views, and invoke codex exec to launch sub-agents. These actions are consistent with the skill's purpose as a development orchestrator.
[EXTERNAL_DOWNLOADS]: The README.md file contains installation instructions that reference a repository hosted on GitHub. This is a standard method for distributing and installing AI agent skills.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-defined configurations from a plan.yaml file and renders them into prompt templates for sub-agents. * Ingestion points: The scripts/spawn_agents.sh script parses user-editable fields such as goal, role, and paths from the plan.yaml file. * Boundary markers: The extracted data is interpolated into the assets/templates/agent-task.md template using sed. The template does not utilize explicit delimiters or specialized instructions to prevent the sub-agents from interpreting user-provided text as system-level commands. * Capability inventory: The sub-agents launched by this skill have the authority to perform git operations and modify files within their designated paths. * Sanitization: The skill employs escaping for sed-specific characters to ensure the technical integrity of the shell command during template rendering, but it does not perform semantic sanitization of the content to mitigate prompt injection risks.

multi-codex-orchestrator