Lov Bp

BP 工坊 · BP Studio

Compose only the stages the user needs. The kit preserves the existing $lov-bp entrypoint while exposing three focused skills that can be used independently.

Triggers

Activate for requests containing or clearly implying:

• Chinese: “商业计划书”, “BP 大纲”, “融资 PPT”, “路演材料”, “投资人演示文稿”, “BP 审稿”, “BP 润色”, or “从项目材料做完整 BP”;
• English: “business plan”, “BP outline”, “pitch deck”, “investor deck”, “fundraising deck”, “review my BP”, or “polish this deck”.

Do not activate for a generic company introduction, ordinary product presentation, annual report, marketing proposal, or document formatting request unless the user also needs an investor-facing business plan.

Runtime and portability

Use Python 3.8+ for workspace and audit scripts. Resolve presentation generation through the active Agent Skills environment; never assume an author's private path. Take user-specific output, brand, and asset paths from the request or current project.

Kit Map

Default full pipeline:

project evidence
      ↓
bp-outline ── evidence gate ──→ bp-deck ── visual gate ──→ bp-polish
      │                             │                         │
      └ brief / ledger / outline    └ PPTX / PDF / preview   └ report / fixes

Routing Rules

Select the smallest route that fulfills the request:

Do not run all modules merely because they exist. For example, a user asking for an outline should not wait for image generation.

For the state machine and handoff semantics, read references/composition.md.

Shared Workspace Contract

All modules read and write the same portable workspace:

business-plan/
├── brief.md
├── evidence-ledger.md
├── outline.md
├── assets/
├── deck-manifest.md
├── reports/bp-review.md
├── project-bp.pptx
├── project-bp.pdf
└── project-bp-preview.png

Rules:

• Never recreate a file that already contains accepted user work.
• Preserve evidence IDs across modules.
• outline.md is the narrative source of truth.
• deck-manifest.md records style, slide count, assets, filenames, and render status.
• reports/bp-review.md records findings and fixes; it must not silently rewrite facts.
• A later module may send work back to an earlier module when it finds a blocker.

Orchestration Workflow (MANDATORY)

Step 0: Resolve the kit

Resolve this SKILL.md directory as KIT_DIR. Module entrypoints are:

$KIT_DIR/skills/bp-outline/SKILL.md
$KIT_DIR/skills/bp-deck/SKILL.md
$KIT_DIR/skills/bp-polish/SKILL.md

Before executing a selected module, read that module's SKILL.md completely and follow its references. If a standalone installation is being used, resolve its own directory as SKILL_DIR.

Verify every selected module before execution:

1. read kit.yaml and resolve the declared relative module path inside KIT_DIR;
2. confirm the selected module's SKILL.md exists;
3. if it is missing, stop before producing partial output and report: BP Skill Kit installation is incomplete: missing <relative path>;
4. tell the user to reinstall the self-contained lov-bp package or install the requested standalone module. Do not silently redirect to a sibling directory or improvise a replacement workflow.

Step 1: Inspect context before asking

Inspect the current project, previous BP workspace, supplied deck, conversation, and links. Reuse known audience, financing stage, ask, language, brand, and output path.

Ask at most one compact round of questions only when an answer changes the result. If the user says “不要问”“直接做”“按推荐方案”, use these defaults:

• seed-stage investors who may not understand the technology;
• 12–15 slides / 8–10 minutes;
• same language as the user;
• clean editorial style with one brand color;
• PPTX + PDF + preview + review report.

Step 2: Run the selected module(s)

For each module:

1. announce the module and why it is needed;
2. read its full SKILL.md;
3. pass the existing workspace instead of starting over;
4. verify its output contract;
5. continue to the next module only if requested or implied by the chosen route.

Step 3: Respect the two useful gates

Evidence gate — after bp-outline:

• product definition is investor-readable;
• all 12 investor questions are covered;
• core numbers have sources or are labeled assumptions;
• no fabricated traction, market, quote, or revenue.

If the user only asked for an outline, stop here. In an interactive full workflow, show the concise page map and ask whether to produce PPT. Skip this pause when the user already requested a complete deck or autonomous execution.

Visual gate — after bp-deck:

• PPTX/PDF/preview exist and page counts match;
• every page has one conclusion and a legible proof;
• logos, images, sources, and QR codes survive final rendering.

Then use bp-polish for the final adversarial review.

Step 4: Handle loop-backs

bp-polish may return one of three verdicts:

• copy fix → patch the outline and regenerate only affected slides;
• evidence blocker → return to bp-outline, update the ledger, then regenerate;
• visual defect → return to bp-deck, regenerate affected slides, and recheck.

Limit automatic correction to three cycles. If a core source remains unavailable, leave an explicit gap rather than guessing.

Step 5: Deliver

Return clickable paths to all artifacts created by the selected route. For a full pipeline, report:

• outline and evidence ledger;
• editable PPTX and PDF;
• full-deck preview;
• deck manifest;
• final review report;
• at most five unresolved decisions or evidence gaps.

Composition Examples

$lov-bp-outline 根据当前仓库和用户访谈写融资大纲

$lov-bp-deck ./business-plan/outline.md --style minimal

$lov-bp-polish ./business-plan/project-bp.pdf --full

$lov-bp 从当前项目材料生成完整种子轮 BP，不要问，按推荐方案

Backward Compatibility

The original commands remain available:

python3 "$KIT_DIR/scripts/init_bp.py" --name "Project" --output ./business-plan
python3 "$KIT_DIR/scripts/audit_bp.py" --input ./business-plan/outline.md

They delegate to the same templates and rubric used by the child skills.

Non-Negotiables

• Do not conflate “write an outline” with “produce a finished deck”.
• Do not choose a visual style before the investment narrative is coherent.
• Do not rewrite facts during visual polishing.
• Do not force a user through all modules when one module solves the request.
• Do not finish a full route without PPTX, PDF, preview, and review report.

Runtime context (shared)

运行前读取本 Skill 包的 skill.yaml，由宿主提供 skill-runtime/v1 上下文。字段解析顺序为：当前请求、项目上下文、个人 Preferences、品牌 Profile、通用默认值。

• 只使用 Manifest 声明的字段；Profile 保存公开品牌事实，Preferences 保存个人工作偏好。
• required: true 字段缺失时，按 Manifest 的问题配置向用户提出一个聚焦问题；用户明确同意后再保存回答。
• 报错提供可复制的 context_id、字段路径与来源，诊断内容避开秘密、完整私人路径和原始配置。

通用反馈闭环

用户在 Skill 驱动任务中提出修改意见时，继续当前产物前必须执行：

1. 先判断意见是 task-specific（仅本次）还是 reusable（可跨任务复用）。
2. task-specific 只修改当前任务，不改 Skill。
3. reusable 先确定作用域：领域规则先更新对应 canonical Skill；适用于所有 Skill 的规则先更新共享规范。
4. 完成规则更新、版本、lint 与分发核验后，再把修改应用到当前任务。
5. reusable 修改会使此前的“确认”“继续”“发吧”失效；完成当前产物修改和回读后必须停下，等待用户下一步指示，不自动进入发布、提交或其他外部写入。