AgentSkillsCN

leanspec-sdd

在 AI 辅助开发中采用 Spec-Driven Development 方法论。当您在 LeanSpec 项目中开展工作时,可使用此方法。

SKILL.md
--- frontmatter
name: leanspec-sdd
description: Spec-Driven Development methodology for AI-assisted development. Use when working in a LeanSpec project.
compatibility: Requires lean-spec CLI or @leanspec/mcp server
metadata:
  author: LeanSpec
  version: 0.2.23
  homepage: https://leanspec.dev

LeanSpec SDD Skill

Teach agents how to run Spec-Driven Development (SDD) in LeanSpec projects. This skill is an addon: it does not replace MCP or CLI tools.

When to Use This Skill

Activate this skill when any of the following are true:

  • The repository contains a specs/ folder or .lean-spec/config.json
  • The user mentions LeanSpec, specs, SDD, or spec-driven planning
  • The task requires multi-step changes, breaking changes, or design decisions

Core Principles

  1. Context Economy: Keep specs under 2000 tokens when possible. Split large specs.
  2. Discovery First: Always run board/search before creating new specs.
  3. Intent Over Implementation: Capture why first, then how.
  4. Progressive Disclosure: Keep SKILL.md concise; use references for details.
  5. No Manual Frontmatter: Use tools to update status, tags, dependencies.
  6. Verify Against Reality: When asked about completion or progress, check the actual codebase, commits, and changes—not just the spec status.

Core SDD Workflow

1) Discover

  • Get the project state: run board (or lean-spec board).
  • Search for related work before creating anything: search (or lean-spec search "query").

2) Design

  • If a spec is needed, create it with create (or lean-spec create).
  • Prefer standard templates and keep scope clear.
  • Validate token count using tokens (or lean-spec tokens).

3) Implement

  • If draft is enabled, move draftplanned once the spec is reviewed, then set in-progress before coding.
  • Skipping planned from draft requires a --force override (CLI/MCP/UI).
  • Document decisions and progress inside the spec as work happens.
  • Link dependencies using link/unlink as they are discovered.

4) Validate & Complete

  • Run validate (or lean-spec validate) before completion.
  • Ensure all checklist items are checked.
  • Verify actual implementation: When asked about completion status or progress:
    • Check git commits and file changes
    • Review actual code implementations
    • Verify test coverage and results
    • Don't rely solely on spec status field
  • Update status to complete only when both spec criteria and actual implementation are verified.

Tool Reference

Use MCP tools when available. Use CLI as fallback.

ActionMCP ToolCLI Command
Project statusboardlean-spec board
List specslistlean-spec list
Search specssearchlean-spec search "query"
View specviewlean-spec view <spec>
Create speccreatelean-spec create <name>
Update statusupdatelean-spec update <spec> --status <status>
Dependenciesdepslean-spec deps <spec>
Relationshipsrelationshipslean-spec rel <spec>
Link / unlink (deprecated)link / unlinklean-spec link/unlink <spec> --depends-on <other>
Token counttokenslean-spec tokens <spec>
Validatevalidatelean-spec validate

Best Practices (Summary)

  • Keep AGENTS.md project-specific only; put SDD methodology here.
  • Never create spec files manually; use create.
  • Keep specs short and focused; split when >2000 tokens.
  • Always check dependencies and link specs that block each other.
  • Document trade-offs and decisions as they happen.

Choosing Relationship Type

Parent/Child = Decomposition (organizational)

  • "This spec is a piece of that umbrella's scope"
  • Spec doesn't make sense without the parent context
  • Parent completes when all children complete

Depends On = Blocking (technical)

  • "This spec needs that spec done first"
  • Specs are independent work items
  • Could be completely unrelated areas

Rule: Never use both parent AND depends_on for the same spec pair.

Test: If the other spec didn't exist, would your spec still make sense?

  • NO → Use parent (it's part of that scope)
  • YES → Use depends_on (it's just a blocker)

See detailed guidance in:

Compatibility Notes

  • Works with any Agent Skills-compatible tool (Claude, Cursor, Codex, Letta, Factory).
  • Requires either LeanSpec MCP tools or CLI access to manage specs.
  • This skill is additive and does not change existing LeanSpec tooling.