Top Codex Coder's Plan.md Template (github.com)

Aditya Advani released a lightweight workflow for AI-assisted coding that standardizes task-scoped work plans called "ExecPlans" via two markdown templates: .agent/template/PLAN.md and AGENTS.md. The AGENTS.md tells your AI agent to instantiate a new ExecPlan per task (e.g., .agent/orchestrator/PLAN.md), map one ExecPlan to one task (feature, refactor, bugfix), and then run it with the command-like prompt “Do the plan.” Plans are living documents that capture purpose, context, TDD-first test cases, step-by-step implementation, decision logs, progress, and verification steps; typically an agent completes a plan in one shot or 2–3 iterations. Copy the repo’s .agent/template/PLAN.md and AGENTS.md into your project (optionally add .agent/* !.agent/template/ to .gitignore to keep plans local) and delete per-feature folders after merge if desired. The templates are MIT-licensed. Significance: this creates repeatable, auditable workflows for code-generation agents, enforcing structured thinking, test-driven development, and traceable decisions—addressing common failure modes like context drift and undocumented changes. Practically, it makes multi-step features easier to manage, supports per-feature isolation, and encodes verification steps so agents can be more reliable. Concurrency is possible but requires orchestration; the author invites community feedback on running multiple concurrent agents (e.g., Codex web or codex CLI background jobs) to evolve the workflow.