Textum: PRD→Scaffold→Story workflow

VerifiedSafe

This skill implements a multi-stage workflow that transforms a product requirements document (PRD) into a scaffold (global context) and then into executable stories, with low-noise outputs and mandatory gate checks at each stage. It supports 16 distinct stages from PRD planning through story execution, and accepts Chinese commands for routing between stages. Useful when you need a disciplined, low-noise progression from high-level requirements to concrete, executable tasks.

Sby Skills Guide Bot
DevelopmentIntermediate
1106/2/2026
Codex
#prd#scaffold#story#workflow#low-noise

Recommended for

Our review

Textum is a structured workflow to go from PRD to code through planning, rendering, validation and slicing stages, with a strong emphasis on low-noise outputs.

Strengths

  • Reduces noise by avoiding unnecessary narration across stages.
  • Multi-window: each stage is self-contained.
  • Includes gate checks to validate quality at each phase.
  • Supports Chinese intent routing.

Limitations

  • Requires `uv` installed and a one-time sync.
  • The experimental `Story Full Exec` may be unstable.
  • Failure tracking is text-based and can be verbose.
When to use it

When you need to generate code from a PRD in a repeatable, quality-controlled manner.

When not to use it

For very simple projects or when traceability from PRD to code is not needed.

Security analysis

Safe
Quality score92/100

The skill is a declarative workflow description with no dangerous commands; only uses 'uv sync' for local Python dependency management, which is safe and standard.

No concerns found

Examples

PRD Plan
需求澄清,帮我生成PRD计划。
PRD Render
输出PRD,基于现有的参考资料。
Story Exec
Story Exec 执行当前story。

name: textum description: Textum PRD→Scaffold→Story workflow for Codex with low-noise outputs and gate checks.

Textum

Hard constraints:

  • Low-noise is non-negotiable (avoid attention/context pollution).
  • Multi-window: each stage is self-contained; do not narrate upstream/downstream flow.
  • Output “next step” as a stage name only.

Prereq (runtime):

  • uv installed.
  • Run uv sync --project .claude/skills/textum/scripts once (creates .claude/skills/textum/scripts/.venv).

Supported stages:

  • PRD Plan → references/prd-plan.md
  • PRD Check → references/prd-check.md
  • PRD Render → references/prd-render.md
  • PRD Slice → references/prd-slice.md
  • Scaffold Plan → references/scaffold-plan.md
  • Scaffold Render → references/scaffold.md
  • Scaffold Check → references/scaffold-check.md
  • Split Plan → references/split-plan.md
  • Split Generate → references/split.md
  • Split Check1 → references/split-check1.md
  • Split Check2 → references/split-check2.md
  • Split Checkout → references/split-checkout.md
  • Story Check → references/story-check.md
  • Story Pack → references/story-pack.md
  • Story Exec → references/story.md
  • Story Full Exec (experimental) → references/story-full-exec.md

Routing:

  • CN intent examples:
    • PRD Plan: 需求澄清 / 澄清需求 / PRD 计划
    • PRD Render: 生成PRD / 渲染PRD / 输出PRD
    • PRD Check: 校验PRD / 检查PRD / 门禁
    • PRD Slice: PRD 切片 / 切片 / 低噪切片 / slice
    • Scaffold Plan: 上下文提取 / 全局上下文 / Scaffold 计划
    • Scaffold Render: 生成GLOBAL-CONTEXT / 渲染GLOBAL-CONTEXT / 输出GLOBAL-CONTEXT
    • Scaffold Check: 校验GLOBAL-CONTEXT / 检查GLOBAL-CONTEXT / GC 门禁
    • Split Plan: Story 拆分规划 / Split Plan / 拆分计划
    • Split Generate: 生成Story / Split Generate / 拆分生成
    • Split Check1: Split 校验1 / 拆分校验1 / 结构阈值校验
    • Split Check2: Split 校验2 / 拆分校验2 / 引用一致性校验
    • Split Checkout: Split Checkout / 依赖图 / 导出依赖图
    • Story Check: Story 校验 / Story Check / 单 Story 门禁
    • Story Pack: Story 执行包生成 / Story Pack / 生成执行包
    • Story Exec: Story 执行 / Story Exec / 单 Story 执行
    • Story Full Exec: Story 批量执行 / Story Full Exec / 试验性全执行
  • If intent is unclear, ask the user to pick one: PRD Plan / PRD Check / PRD Render / PRD Slice / Scaffold Plan / Scaffold Render / Scaffold Check / Split Plan / Split Generate / Split Check1 / Split Check2 / Split Checkout / Story Check / Story Pack / Story Exec / Story Full Exec.

Always:

  • For every FAIL item (in diagnostics/replan packs): include loc/problem/expected/impact/fix, and fix must be a single action.
  • Keep chat output low-noise: prefer paths (wrote: docs/*) over pasting long FAIL lists.
Related skills