Token导航 LogoToken导航TokenDH.com
研究检索只读github未标认证来源可访问许可证需确认审计通过

sdd-workflowSDD 工作流程

Agent Skill

sdd-workflow 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

259

周安装

11

GitHub Stars

公开资料未说明

下载量

91
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

unknown

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:sdd-workflow(SDD 工作流程)
来源仓库:https://github.com/9ashwin/on-my-sdd
仓库路径:skills/sdd-workflow
安装命令:
npx skills add https://github.com/9ashwin/on-my-sdd --skill sdd-workflow
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/9ashwin/on-my-sdd --skill sdd-workflow

简介

sdd-workflow 用于查找、检索和筛选相关信息。

  • 适合在 Codex、Claude、Cursor、Gemini CLI 中根据关键词、任务场景或来源线索快速定位候选结果。
  • 通过 npx skills add 命令从指定 GitHub 仓库安装,需结合原始 README 确认具体用法。
  • 安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。
  • 注意该技能当前分类为研究检索,实际功能以来源仓库文档为准。

SKILL.md

IF A SPEC EXISTS, YOU MUST READ IT BEFORE WRITING CODE. IF NO SPEC EXISTS FOR BEHAVIOR CHANGE, YOU MUST CREATE ONE FIRST.

This is not negotiable. This is not optional. You cannot rationalize your way out of this.

This skill IS the SDD pipeline. Reference it once in CLAUDE.md — you do NOT need to list individual steps. The decision tree, phase detection, and transition rules below handle all routing.

Instruction Priority

  1. User's explicit instructions (CLAUDE.md, AGENTS.md, direct requests) — highest priority
  2. OpenSpec artifacts (proposal.md, specs/, design.md, tasks.md) — the authoritative spec baseline
  3. SDD workflow skills — route and enforce process
  4. Default system prompt — lowest priority

If the user says "skip the spec, just write code," follow the user's instructions. The user is in control.

SDD Workflow — Spec-Driven Development Router

The Rule

Before any code, human and AI agree on what to build. Specifications are files in openspec/. Every behavior change is traceable from proposal through archive.

Request Classification

When the user brings a development request, classify FIRST. Then route.

digraph sdd_routing {
    "User request received" [shape=doublecircle];
    "Is it a one-line fix?\n(typo, log line, comment)" [shape=diamond];
    "Make change,\nverify directly" [shape=box];
    "Is it a bug with\nunclear cause?" [shape=diamond];
    "superpowers:systematic-debugging\n(root cause first)" [shape=box];
    "Are there OpenSpec\nartifacts already?" [shape=diamond];
    "Read existing artifacts,\npick up where left off" [shape=box];
    "Is the requirement\nclearly bounded?" [shape=diamond];
    "Unfamiliar codebase\nor uncertain approach?" [shape=diamond];
    "/opsx:propose\n(generate 4 artifacts)" [shape=box];
    "/opsx:explore\n(build context first)" [shape=box];
    "superpowers:brainstorming\n(Socratic design)" [shape=box];
    "Route to execution\n(sdd-review-specs →\nwriting-plans → apply)" [shape=doublecircle];

    "User request received" -> "Is it a one-line fix?\n(typo, log line, comment)";
    "Is it a one-line fix?\n(typo, log line, comment)" -> "Make change,\nverify directly" [label="yes"];
    "Is it a one-line fix?\n(typo, log line, comment)" -> "Is it a bug with\nunclear cause?" [label="no"];
    "Is it a bug with\nunclear cause?" -> "superpowers:systematic-debugging\n(root cause first)" [label="yes"];
    "Is it a bug with\nunclear cause?" -> "Are there OpenSpec\nartifacts already?" [label="no"];
    "Are there OpenSpec\nartifacts already?" -> "Read existing artifacts,\npick up where left off" [label="yes"];
    "Are there OpenSpec\nartifacts already?" -> "Is the requirement\nclearly bounded?" [label="no"];
    "Is the requirement\nclearly bounded?" -> "Unfamiliar codebase\nor uncertain approach?" [label="yes, but\nneeds exploration"];
    "Is the requirement\nclearly bounded?" -> "/opsx:propose\n(generate 4 artifacts)" [label="yes, clear\nboundaries"];
    "Unfamiliar codebase\nor uncertain approach?" -> "/opsx:explore\n(build context first)" [label="need to read\ncode first"];
    "Unfamiliar codebase\nor uncertain approach?" -> "superpowers:brainstorming\n(Socratic design)" [label="greenfield\nor approach\ncomparison"];
    "/opsx:propose\n(generate 4 artifacts)" -> "Route to execution\n(sdd-review-specs →\nwriting-plans → apply)";
    "/opsx:explore\n(build context first)" -> "Is the requirement\nclearly bounded?";
    "superpowers:brainstorming\n(Socratic design)" -> "Is the requirement\nclearly bounded?";
    "Make change,\nverify directly" -> "Route to execution\n(sdd-review-specs →\nwriting-plans → apply)" [style=invis];
}

Phase Detection

Check the file system to determine where you are in the workflow:

What existsPhaseNext action
No openspec/ directoryUninitializedRun openspec init first
openspec/ exists, no change dirReady for proposal/opsx:propose <name> or exploration
openspec/changes/<name>/ with 4 artifacts, unreviewedSpecs need reviewTier 1: self-review inline (2-5 min). Tier 2: sdd-review-specs full gate. Default Tier 2 if unsure.
openspec/changes/<name>/ with reviewed artifactsReady for executionsuperpowers:writing-plans
tasks.md has unchecked itemsIn progress/opsx:apply + superpowers:test-driven-development
All tasks checked, not archivedReady for deliverysuperpowers:verification-before-completion/opsx:archive

Tool Selection Matrix

When both OpenSpec and Superpowers offer a tool for the same phase, use this:

ScenarioUse ThisNot ThatWhy
Reading existing code, finding patterns/opsx:explore@brainstormingExplore reads code; brainstorming generates ideas
Defining new feature from scratch@brainstorming/opsx:exploreBrainstorming compares approaches; explore describes existing state
Generating spec artifacts/opsx:propose@writing-plansPropose creates the 4-artifact structure; writing-plans refines granularity
Refining task granularity@writing-plansManual onlyWriting-plans converts coarse tasks to 2-5min bite-sized units
Executing tasks/opsx:apply + @test-driven-developmentEither aloneApply is the scheduler; TDD is the executor. Pipeline them.
Debugging failures@systematic-debuggingDirect fixesRoot cause investigation first. Never trial-and-error.
Code review@requesting-code-review + @receiving-code-review"Looks good to me"Structured review with independent context
Claiming completion@verification-before-completion"Should work now"Fresh verification evidence required
Archiving completed work/opsx:archiveManual file movesArchive does delta merge + timestamp + project.md update

Red Flags

These thoughts mean STOP — you're rationalizing skipping the SDD process:

ThoughtReality
"This is simple, I don't need a spec"Simple changes cause complex bugs. A 5-line proposal.md saves hours.
"I'll write the spec after the code"Specs-after describe what you built, not what's needed. They don't align.
"The spec is in the conversation history"Conversation history evaporates. Files persist. Write it down.
"I already know what to build"Knowing ≠ having it reviewed. Specs are the agreement, not the idea.
"Specs slow me down"Rework from misaligned expectations is slower. Align first, execute second.
"This is just a prototype"Prototypes become production. Spec now saves migration pain later. "Prototype" is a scope label in proposal.md — it doesn't exempt you from writing specs/, design.md, and tasks.md.
"I'll just explore the codebase first"Use /opsx:explore — structured exploration with output. Don't browse aimlessly.
"Let me check a few files quickly"That's what /opsx:explore does systematically. Use the tool.
"I remember how this codebase works"Code evolves. Your memory is stale. Read the specs.
"One more task before reviewing"Review gate exists for a reason. Skipping it accumulates unchecked assumptions.
"The tasks.md checklist IS the review"tasks.md tracks implementation progress. sdd-review-specs validates scope, design, and completeness. These are different activities — checking boxes is not reviewing specs. The review gate has a 7-step function (identify tier, read artifacts, check against checklist, communicate issues, iterate, declare pass, route). Skipping it is non-compliance.
"I'll review as I implement"Review-after-implementation finds what you built, not what you need. The gate must be BEFORE code — catching scope and design issues after implementation costs 10x more to fix.
"The real risk is the deadline, not the spec"Risk inversion. Skipping the spec *is* the deadline risk — unreviewed changes cause the incidents that blow deadlines.
"A minimal proposal is good enough"Partial compliance is non-compliance. The SDD pipeline produces 4 artifacts for a reason. A proposal without specs/design/tasks is a wish, not a plan.
"Can't spec what I don't understand yet"That's what /opsx:explore is for. Build understanding first, then spec. Never invert the order — code is not exploration.

All of these mean: follow the SDD process. No shortcuts.

Skill Types

SDD workflow is a RIGID skill. Follow the routing exactly. Don't adapt away the sequence.

The tools it routes to have their own types:

  • @test-driven-development, @systematic-debugging, @verification-before-completionRigid. Follow exactly.
  • @brainstorming, @writing-plans, /opsx:exploreFlexible. Adapt principles to context.
  • /opsx:propose, /opsx:apply, /opsx:archiveRigid. These are CLI tools with defined behavior.

Transition Rules

After each phase completes, route to the next:

Phase complete → check what's next:

/opsx:propose done      → DETERMINE TIER, then route:
  Tier 0 (typo/log/comment)   → skip /opsx:propose entirely — make change and verify directly
  Tier 1 (single field, config)→ self-review inline: check proposal.md scope boundary
                                 and tasks.md executability. Skim design.md for obvious
                                 conflicts with existing architecture. 2-5 min.
                                 Then proceed to implementation.
  Tier 2 (new feature, refactor)→ HARD STOP. Invoke sdd-review-specs full gate function.
                                 Produce review.md. Do NOT proceed until it passes.
                                 Self-review via tasks.md checkboxes is NOT review for Tier 2.
  When in doubt, default to Tier 2. Over-reviewing is cheaper than missing a critical issue.
sdd-review-specs passed → superpowers:writing-plans (refine task granularity)
writing-plans done      → /opsx:apply + superpowers:test-driven-development (pipeline)
                          OR superpowers:subagent-driven-development (parallel tasks)
                          OR superpowers:executing-plans (batch execution)

During apply:
  - Error/failure → superpowers:systematic-debugging → return to apply
  - All tasks done → superpowers:verification-before-completion

verification-before-completion passed → /opsx:archive
/opsx:archive done → commit → push → done

Related Skills

  • sdd-review-specs — Structured review of OpenSpec 4 artifacts before implementation
  • superpowers:brainstorming — Socratic design for greenfield features
  • superpowers:writing-plans — Convert coarse tasks to 2-5min bite-sized units
  • superpowers:test-driven-development — RED-GREEN-REFACTOR cycle
  • superpowers:systematic-debugging — Root cause investigation before fixes
  • superpowers:verification-before-completion — Evidence before completion claims
  • superpowers:requesting-code-review — Structured code review
  • superpowers:finishing-a-development-branch — Merge/PR/keep/discard decisions

OpenSpec command reference: /opsx:explore, /opsx:propose, /opsx:apply, /opsx:verify, /opsx:archive

The Bottom Line

Specs before code. Review before execution. Evidence before completion.

The SDD process exists because AI programming without a spec layer drifts. Each phase produces a file in openspec/ — that file is the agreement. Honor it.

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Codex

32.5%
按下载量换算30

Claude

32.69%
按下载量换算30

Cursor

20.71%
按下载量换算19

Gemini CLI

9.78%
按下载量换算9

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

只读

该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills