Work with Obsidian vaults (plain Markdown notes) and automate via obsidian-cli.
npx skills add zhenxuanshi-ship-it/business-doc-skill-suite --skill "business-doc-pipeline"
Install specific skill from multi-skill repository
# Description
Orchestrate a business document workflow across discovery, drafting, review, revision, and finalization. Use when the user wants an end-to-end workflow for 项目可行性研究报告, 项目建议书, PRD, 产品需求说明书, 架构设计, 技术方案, 实施方案, or wants to move from notes to draft to review to revision. Triggers on requests like: 从头到尾搞定这份文档, 帮我走完整流程, 先写再 review 再修改, end-to-end business document workflow, document pipeline, 从需求到成稿, 把这份方案一路做完.
# SKILL.md
name: business-doc-pipeline
description: Orchestrate a business document workflow across discovery, drafting, review, revision, and finalization. Use when the user wants an end-to-end workflow for 项目可行性研究报告, 项目建议书, PRD, 产品需求说明书, 架构设计, 技术方案, 实施方案, or wants to move from notes to draft to review to revision. Triggers on requests like: 从头到尾搞定这份文档, 帮我走完整流程, 先写再 review 再修改, end-to-end business document workflow, document pipeline, 从需求到成稿, 把这份方案一路做完.
Business Doc Pipeline
This skill is a lightweight orchestrator.
It does not replace business-doc-writer or business-doc-reviewer.
Its job is to:
- detect current stage,
- recommend the next stage,
- route to the right skill,
- keep the workflow structured,
- avoid messy jumps.
Agent execution rule
This skill is a lightweight orchestrator.
It does not replace business-doc-writer or business-doc-reviewer.
Its job is to:
- detect current stage,
- recommend the next stage,
- route to the right skill,
- keep the workflow structured,
- avoid messy jumps.
Default: main agent does the heavy work directly. Writing, review, revision, and other substantial tasks are done in the main session by default.
Sub-agent mode only when user specifies. Only spawn a sub-agent when the user explicitly asks for it, such as:
- "用子代理写"
- "派生子代理重跑"
- "spawn child agent"
Core stages
- discovery
-
clarify goals, audience, scope, constraints, and missing inputs
-
blueprint
-
choose doc type and outline
-
draft
-
generate first usable draft
-
review
-
run multi-role document review
-
revise
-
revise based on issues and roadmap
-
finalize
- prepare the final usable version
Routing rules
Start at discovery when:
- the user has only ideas, notes, or vague goals
- the user says they are not sure what kind of doc to write
- the scope is still unclear
Start at blueprint when:
- the goal is clear but the structure is not
- the user asks for an outline first
Start at draft when:
- the user already knows the doc type and wants a first draft
- enough context exists to write a useful version
Start at review when:
- the user already has a draft
- the user asks to critique, challenge, or assess a document
Start at revise when:
- the user has review comments or a revision roadmap
- the user asks to update an existing draft based on feedback
Start at finalize when:
- the draft is already good enough
- the user mainly wants a clean final version or delivery package
Pipeline behavior
Step 1: detect current stage
Infer from user input and materials.
Common cues:
- notes / minutes / rough ideas -> discovery
- outline request -> blueprint
- "write" -> draft
- "review" / "评审" -> review
- "revise" / "修改" / comments -> revise
- "final version" / "定稿" -> finalize
Step 2: choose document type
Map the work to one of these:
- feasibility-report
- project-proposal
- prd
- architecture-design
- implementation-plan
- executive-summary
Use references/doc-stage-matrix.md when the type is unclear.
Step 3: produce a checkpointed recommendation
After deciding the stage, explicitly tell the user:
1. current stage
2. target doc type
3. recommended next action
4. what input is still missing
Use concise checkpoint language.
Step 4: hand off to the right skill
- drafting work ->
business-doc-writer - review work ->
business-doc-reviewer
Step 5: keep the workflow honest
Do not pretend the pipeline is complete if major gaps remain.
Always surface:
- assumptions
- blockers
- missing inputs
- next best action
Recommended flow patterns
Pattern A: from messy notes to usable draft
discovery -> blueprint -> draft
Use when the user is early-stage.
Pattern B: draft to decision-ready document
draft -> review -> revise -> finalize
Use when the user already has something written.
Pattern C: full workflow
discovery -> blueprint -> draft -> review -> revise -> finalize
Use when the user wants end-to-end help.
Checkpoint format
Use this style when helpful:
## Pipeline checkpoint
- Current stage: ...
- Document type: ...
- Current status: ...
- Main gap: ...
- Recommended next step: ...
Revision loop rule
Keep revision loops practical.
Default:
- 1 review round
- 1 main revision round
- optional final polish
Do not create endless loops unless the user explicitly wants multiple rounds.
Output expectations by stage
discovery
Output:
- concise brief
- assumptions
- open questions
- recommended doc type
blueprint
Output:
- outline
- section purpose
- missing input list
draft
Output:
- usable first draft
- assumptions/open questions section if needed
review
Output:
- overall judgment
- prioritized issue list
- revision roadmap
revise
Output:
- revised content or a clear revision plan
- explanation of major changes
finalize
Output:
- cleaned final draft
- optional one-page summary if useful
References
Read when relevant:
- references/doc-stage-matrix.md — map situation to stage and doc type
- references/checkpoint-patterns.md — checkpoint message patterns
- references/handoff-protocol.md — how to move between writer and reviewer
# Supported AI Coding Agents
This skill is compatible with the SKILL.md standard and works with all major AI coding agents:
Learn more about the SKILL.md standard and how to use these skills with your preferred AI coding agent.