manhua-man/codex-parallel-subagent-planner

6 stars · Last commit 2026-08-05

Parallel subagent planner skill for Codex/Claude Code: artifact-first lane planning, explorer/worker orchestration, model+reasoning allocation

README preview

# parallel-subagent-planner (v3.0.1)

[English](README.md) | [简体中文](README.zh-CN.md)

`parallel-subagent-planner` is a lightweight **Agent Planning Harness Skill** for Codex. It helps Codex decide when to split work into subagents, set safe file scope boundaries, establish execution order, generate clean subagent prompts, and recommend reusable custom agent roles.

---

## What It Does

- **Split Decision**: Evaluates whether splitting into subagents saves wall-clock time vs. direct execution, or if a read-only investigation de-risks implementation.
- **File Boundary Isolation**: Enforces disjoint write scopes (`write(A) ∩ write(B) = ∅`) so parallel subagents never overwrite each other's work.
- **Execution Order**: Assigns shared contract files to exactly one owner (main thread or subagent) before dependent consumers read them.
- **Clean Subagent Prompts**: Generates clear prompts specifying goals, read/write file scopes, and acceptance tests.
- **Main Thread Integration**: Merges subagent outputs and runs workspace-wide integration checks.
- **Custom Agent Guidance**: Suggests saving recurring subagent roles into `.codex/agents/<name>.toml` specs with explicit user approval when repeated evidence exists.

---

## Core Cycle

View full repository on GitHub →