plans-methodology（计划方法）

Plans Methodology

Structured work tracking for individuals or multi-agent teams. Each agent maintains plans/ in their workspace.

Directory Structure

plans/
  README.md
  draft/          # Being defined, not yet approved
  approved/       # Ready for execution
  in_progress/    # Currently executing
  completed/      # Done
  archived/       # Historical reference

Each plan is a folder: YYYY-MM-DD-plan-name/

Plan Contents

YYYY-MM-DD-plan-name/
  specs/              # Problem statements, requirements
  proposal.md         # Why: justification, impact, risks
  design.md           # How: architecture, flows, decisions
  tasks.md            # What: checkboxes, owners, phases

proposal.md

# Proposal: [Plan Name]

## Status
Draft | Submitted | Approved | Rejected

## Parent Plan
[agent]/plans/[state]/[plan-name] (if child plan)

## Problem
What problem are we solving?

## Proposed Solution
High-level approach.

## Impact
What changes? What improves?

## Cost / Effort
Time, resources, dependencies.

## Risks
What could go wrong?

## Decision
Approved by [name] on [date] | Rejected because [reason]

design.md

# Design: [Plan Name]

## Overview
What are we building/doing?

## Architecture / Approach
How does it work? Diagrams, flows, decisions.

## Dependencies
What do we need? Other plans, external services, other agents.

## Constraints
Limitations, rules, non-negotiables.

## Open Questions
Things still to be decided.

tasks.md

# Tasks: [Plan Name]

## Summary
Brief overview.

## Tasks

### Phase 1: [Name]
- [x] Task description -> Owner
  Completed YYYY-MM-DD. Notes.
- [ ] Task description -> Owner
  Details, acceptance criteria.
- [ ] Delegate to [Agent]: [what] -> [Agent]
  Child plan: [agent]/plans/[state]/[plan-name]

Lifecycle

1. Draft - Specs written, proposal drafted
2. Approved - Proposal approved, design and tasks written
3. In Progress - Execution started
4. Completed - All tasks done
5. Archived - Kept for reference

Move the plan folder between state directories to change state.

When to Create a Plan

Create when: multi-agent coordination needed, significant effort, multiple phases, needs approval, or progress tracking desired.

Skip for: quick tasks, simple lookups, routine operations.

Approval Authority

Define your own hierarchy. Typical pattern:

• Owner/CEO approves strategic plans
• Lead agent can approve plans aligned with established vision
• Sub-agents submit proposals to their lead for approval

Document your approval chain in your workspace's plans/README.md.

Delegation

When delegating work to another agent:

1. Parent plan's tasks.md references the child plan path
2. Child plan's proposal.md references the parent plan path

Example in parent tasks.md:

- [ ] Delegate to [Agent]: implement feature X
  Child plan: [agent]/plans/in_progress/2026-02-20-feature-x

Example in child proposal.md:

## Parent Plan
[lead-agent]/plans/in_progress/2026-02-20-parent-plan

On Session Start

1. Check plans/in_progress/ for active plans
2. Check plans/draft/ for plans awaiting action
3. Continue where you left off

Granularity Tips

• Not everything needs a plan. Use for significant, multi-step work.
• A single agent working on a quick fix does not need a plan.
• Plans shine when coordination, tracking, or approval is involved.
• Keep plan documents concise. Verbose plans get ignored.