codex-subagents

To run parallel Codex agents, this Skill launches multiple `codex exec` processes concurrently on your current branch. It assigns distinct sub-tasks to individual agents using strict scope prompts, allowing them to work simultaneously and converge efficiently on large feature implementations.

The best way to orchestrate multiple AI agents is by executing them in structured waves with per-agent logging. This approach enforces interactive PTY sessions and provides overseer visibility through git diff and log snapshots, ensuring organized execution and robust monitoring.

Yes, you can use `codex exec` to modify multiple files simultaneously by assigning distinct sub-tasks to individual parallel agents. Each agent operates with explicit scope via prompt files, ensuring targeted code modifications across several files converge into a final implementation.

You monitor parallel subagents and track their code changes through real-time status updates, per-agent logs, and run-centric control. The Skill provides overseer visibility by generating git diff and log snapshots for each agent's execution wave.

Yes, you need a specific manifest file to launch a wave of parallel AI agents. The Skill uses a `.agents/state/subagent-manifest.tsv` file to define and launch the parallel agents, ensuring structured prompt formatting with scope, edits, and outcomes.

You should use strict scope prompts when executing parallel agents to enforce structured prompt formatting and prevent scope creep. This ensures each agent focuses on its specific edits and outcomes, facilitating organized multi-agent wave execution for complex development tasks.