spex

以 /spex 手动调用，按命令路由到 commands/ 下的 SOP，执行 create、modify、apply、apply-one-step、merge、archive、init；配套 Python CLI 脚本与模板，管理规格文档、待办清单、git 提交与合并，覆盖需求分析到增量实现与提交的流程。

jiangxin

热度 11.95

最近更新 2026-08-14 10:33:16

分类：开发依赖：代码执行 / 读写文件应用场景：代码生成 / 工作流 / 任务管理 / 代码审查

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add spex

Metadata

name: spex
disable-model-invocation: true
description: Spec-Driven Development (Spex) skill that manages the full SDLC — from requirement analysis and design to incremental implementation and submission. Invoked manually via /spex <command>. Supports commands: create (new), modify, apply (run, do, go), apply-one-step (step), merge (submit), archive, init.
version: 0.7.0
arguments: [{"name":"command","required":false,"description":"Sub-command to execute. Must be one of: create (alias: new), modify, apply (aliases: run, do, go), apply-one-step (alias: step), merge (alias: submit), archive, init. If omitted, infer intent from the remaining text; route directly when confidence ≥ 90%, otherwise ask user to confirm."},{"name":"prompt","required":false,"description":"Optional context passed to the command. For 'create', this is the requirement describing the spec to generate."}]

Spex — Spec-Driven Development

Usage

/spex [command] [prompt]

IF no args (/spex) -> show Supported Commands table -> STOP
IF recognized command (/spex create ...) -> load matching commands/<file>.md (Command Routing) -> follow that SOP exactly; forward user text as $prompt
IF free-form (/spex <arbitrary text>) -> Free-form Intent Inference

Supported Commands

Command	Aliases	Description
`create`	`new`	Create a spec document (no code changes)
`modify`		Modify a spec's requirements
`apply`	`run`, `do`, `go`	Apply a spec to generate code
`apply-one-step`	`step`	Apply one step from a spec's todo list
`merge`	`submit`	Submit completed work (merge or PR)
`archive`		Archive a completed spec
`init`		Initialize spex environment

Command Routing

Command file paths are relative to this SKILL.md directory.

Match	Command file
`create` / `new`	`commands/create.md`
`modify`	`commands/modify.md`
`apply` / `run` / `do` / `go`	`commands/apply.md`
`apply-one-step` / `step`	`commands/apply-one-step.md`
`merge` / `submit`	`commands/merge.md`
`archive`	`commands/archive.md`
`init`	`commands/init.md`

Routing Discipline

Role: router, not assistant
Resolve command -> load command file -> follow every Phase
User prompt => $prompt for the command SOP only
NEVER act on user prompt directly (no read/write/plan outside SOP)
NEVER skip or shortcut the command SOP
ALWAYS load the full command markdown; follow every Phase as written

Free-form Intent Inference

When first arg matches no route, infer intent:

If the user's text suggests...	Suggest command
A new feature, requirement, or idea to implement	`create`
Changing requirements for an existing spec	`modify`
Starting implementation of a spec	`apply`
Working through a spec one step at a time	`apply-one-step`
Finishing, merging, or submitting completed work	`merge`
Cleaning up completed specs	`archive`
Setting up spex for the first time	`init`

IF confidence >= 90% -> route directly; free-form text => $prompt
IF confidence < 90% OR multiple commands plausible -> ask user to confirm before routing
IF too vague (e.g. "help" / empty) -> show Supported Commands table -> STOP