dotplan

分类: 开发与编程 | 上传者: jamesondh | 下载: 0 | 版本: v1.0（最新）

Use this skill for any non-trivial coding project that spans multiple sessions or phases. dotplan provides structured project memory via a `.planning/` directory convention — spec before code, compact state between sessions, separate implementation from review. Activate when starting a new multi-phase project, continuing work across sessions, or when the user mentions dotplan, `.planning/`, STATE.md, or ROADMAP.md.

更新日志: Source: GitHub https://github.com/jamesondh/dotplan

目录结构

当前层级: 根目录

📁 templates/
- 📄 AGENT-INSTRUCTIONS.md 983 B
- 📄 SPEC.md 644 B
📄 dotplan.png 194.3 KB
📄 dotplan.svg 8.0 KB
📄 init.sh 2.4 KB
📄 LICENSE 1.0 KB
📄 README.md 31.2 KB
📄 SKILL.md 6.1 KB

SKILL.md

---
name: dotplan
description: Use this skill for any non-trivial coding project that spans multiple sessions or phases. dotplan provides structured project memory via a `.planning/` directory convention — spec before code, compact state between sessions, separate implementation from review. Activate when starting a new multi-phase project, continuing work across sessions, or when the user mentions dotplan, `.planning/`, STATE.md, or ROADMAP.md.
---

# dotplan — Project Memory for AI Agents

dotplan is a `.planning/` directory convention and workflow for maintaining continuity across agent sessions. It's git-native, agent-agnostic, and zero-install — just markdown files that travel with the repo.

The core problem: AI agents are stateless. Every new session starts from zero. dotplan solves this with structured state files and workflow conventions.

## Setup

If a project doesn't have a `.planning/` directory yet and needs one (non-trivial, multi-phase work), create it:

```
.planning/
  ROADMAP.md          # ordered phases with status
  STATE.md            # active context — what's happening now
  phases/             # per-phase specs
  _deferred/          # parked ideas not on the roadmap yet
```

Create `.planning/ROADMAP.md`:

```markdown
# Roadmap

## In Progress
- [ ] **Phase 1: {Name}**

## Planned
- [ ] Phase 2: {Name}

## Deferred
```

Create `.planning/STATE.md`:

```markdown
# Project State

## Active Work
{What's currently in progress}

## Blockers
None

## Next Steps
{What comes after current work}
```

Add to `.gitattributes` (create if needed):

```
.planning/** linguist-generated
```

Then add workflow instructions to the project's agent instruction file (`CLAUDE.md`, `AGENTS.md`, `.cursorrules`, etc.) — see the Workflow Instructions section below.

## Workflow Instructions

Add this to the project's agent instruction file alongside existing project context:

### Project Management

- For non-trivial work, start by reading `.planning/STATE.md` and `.planning/ROADMAP.md`.
- For simple tasks (1-2 files, low risk), implement directly without planning ceremony.
- For medium/high-risk work, write `.planning/phases/NN-{name}/SPEC.md` before coding.
- In each task spec, include `Files`, `Docs to update`, `Verify`, and `Done when`.
- After implementation, run verification. For medium/complex work, if no review was performed, ask: "Would you like me to review this with a subagent?"
- At phase wrap-up, update `STATE.md` with a brief completion summary (key changes, verification, issues, follow-ups), then update `ROADMAP.md`.
- Keep `STATE.md` under 150 lines and commit `.planning/` updates with code changes.
- If `STATE.md` exceeds 150 lines, suggest compacting it.

## Complexity Assessment

Assess before starting. Not every task needs the full workflow.

| Level | Signals | Approach |
|-------|---------|----------|
| **Simple** | 1-2 files, clear change, no risk | Just do it. No spec needed. |
| **Medium** | 3-5 files, single-phase, limited risk | Write a spec, implement, review, update STATE.md. |
| **Complex** | Multi-phase, 5+ files, architectural decisions, high-risk (auth, migrations, external APIs, infra) | Full workflow: spec each phase, review spec, implement, review implementation, update STATE.md and ROADMAP.md. |

Risk matters more than file count. A one-file auth change can be higher risk than a ten-file UI refactor.

## Phase Spec Template

Write to `.planning/phases/NN-{name}/SPEC.md` before implementing medium/complex work:

```markdown
# Phase N: {Name}

## Goal
{What this phase accomplishes}

## Risk
{Low / Medium / High. Note migrations, auth changes, external APIs, infra.}

## Tasks

### Task 1: {Description}
**Files:** src/foo.ts (new), src/bar.ts (modify)
**Docs to update:** README.md, CHANGELOG.md — or "None"
**Steps:**
- Step-by-step implementation notes
**Verify:** npm test, go test ./..., etc.
**Done when:** {Concrete completion condition}
```

The **"Docs to update"** field is intentional — documentation is part of the task, not an afterthought.

## The Workflow Loop

**Per phase:**
1. **Spec** — Write `phases/NN-{name}/SPEC.md` with task breakdown.
2. **Implement** — Work through tasks. Update STATE.md as you go.
3. **Review** — Review with a different model or session than implementation. The implementer is blind to their own choices.
4. **Wrap up** — Follow the checklist below.

### Phase Wrap-up Checklist

- Push all commits
- Update `STATE.md` with brief completion summary (key changes, verification, issues, follow-ups)
- Compact STATE.md — keep only active context + brief recent-completed summary
- Update `ROADMAP.md` — mark phase complete, confirm next
- Doc check — did all docs listed in task specs get updated?
- Commit `.planning/` changes
- Review upcoming phases — revise if this phase revealed new complexity

## STATE.md Rules

STATE.md is the bridge between sessions. Keep it focused.

**Keep under 150 lines.** This is a hard constraint.

**Include:**
- Current phase and task-level status
- Last completed phase (brief summary, 5-10 lines)
- Active decisions that affect upcoming work
- Blockers
- Key metrics (test count, build status, whatever matters)

**Exclude:**
- Full history of every past phase
- Verbose logs
- Design rationale that doesn't affect current work (that belongs in the agent instruction file)

## Session Recovery

When picking up a project cold:
1. Read the agent instruction file for stable project context (what, why, stack, conventions)
2. Read `ROADMAP.md` for the big picture
3. Read `STATE.md` for active context
4. If more history is needed, read prior phase `SPEC.md` files

## Key Principles

- **Spec before code** — catches issues early, creates a record of intent vs. result
- **Separate implementation and review** — use different models or sessions; this is the highest-ROI practice
- **State has a size budget** — 150 lines max prevents unbounded context growth
- **Documentation is part of the task** — specs list which docs to update; the agent treats this as required
- **Deferred work has a home** — `_deferred/` prevents scope creep while preserving good ideas
- **Convention over tooling** — no CLI, no install, just markdown files any agent can read

登录后下载/点赞/收藏 ❤ 2 | ★ 0

dotplan

目录结构

SKILL.md

举报内容

提示