Bootstrap Curaye

---

name: curaye-bootstrap description: Bootstrap a new project with a .curaye/ folder. Use when starting a project that has never been tracked under Curaye. Mines the active session context first — if you have been brainstorming, the answers are already in the conversation. Falls back to a targeted interview only for genuine gaps. Generates multiple planned specs from session output, not just one. Optionally pass the project path as an argument. compatibility: Designed for Claude Code. Requires the curaye CLI installed. metadata: version: "0.2"

Bootstrap a new or blank project with a .curaye/ folder. Mines session context first, interviews only for gaps, and produces a complete initial scaffold including all planned specs surfaced in the conversation.

Project path (optional): $ARGUMENTS

If a path is provided, use it. Otherwise use the current working directory.

Step 1 — Pre-flight checks

Check that .curaye/ does not already exist at the target path. If it does, stop: "This project already has a .curaye/ folder. Use /curaye-build to work on existing specs, or /curaye-import if the docs are incomplete."

Scan existing files that can inform the scaffold:

• README.md — project name, description, purpose
• package.json / Cargo.toml / pyproject.toml — stack and app type
• Any loose .md spec or planning files at the project root or in a docs/, specs/, or notes/ folder

Note what you found — you will use this alongside session context in the next step.

Step 2 — Mine the session context

Before asking a single question, extract everything already established in this conversation:

• Project identity — name, one-sentence description, purpose
• Target user — who it is for
• App type — desktop, web app, CLI, mobile, library, other
• Tech stack — languages, frameworks, libraries mentioned or decided
• Planned features — anything described as "we want to build X", "the first thing is Y", "it should do Z"
• Decisions made — any deliberate choices or tradeoffs discussed ("we chose X over Y because…")
• Things ruled out — scope exclusions, non-goals, "not a Z" statements

Present your extraction as a confirmation block before doing anything else:

From our conversation I have:

  Project:  <name> — <one sentence>
  For:      <target user>
  Type:     <app type>
  Stack:    <detected stack, or "not discussed">
  
  Planned specs I found:
    1. <feature or thing to build>
    2. <feature or thing to build>
    ...
  
  Decisions to record:
    - <decision>
    ...
  
  Gaps (I'll ask about these):
    - <anything missing>

Does this look right, or should I adjust anything before we continue?

Wait for confirmation or corrections. Do not proceed until the user confirms.

Step 3 — Fill the gaps (targeted interview)

Ask only about what is genuinely missing after Step 2. Skip any question whose answer was established in the session.

If project identity is missing: "What is this project in one sentence?"

If target user is missing: "Who is it for?"

If app type is missing: Present the choice (Desktop / Web app / CLI / Mobile / Library / Other).

If shared decisions are relevant: Run curaye shared list --category decisions and present results. Ask which apply. Skip silently if the shared layer has no decisions.

If no planned specs were found in session: "What do you want to build first?"

Ask only what is missing. If the session already answered everything, skip this step entirely and say: "I have everything I need — no questions."

Step 4 — Run the mechanical scaffold

curaye init <path>

Step 5 — Draft the documents with AI

Using session context and any interview answers, write:

prd.md — a considered product brief:

• North star from project identity + target user
• "What it does" — from session or interview
• "What it is not" — infer from scope exclusions and ruled-out things discussed in session
• "Target user" — from session or interview

stack.md — from session context first, then shared/stack/ seed for the app type, then detected manifests. Mark unconfirmed entries with # to verify.

product.md — distribution and interface strategy appropriate to the app type.

Planned specs — one spec per feature or thing-to-build surfaced in Step 2. Do not collapse them into one. Each spec gets:

• Problem derived from conversation context
• Goal, non-goals, and 5–8 acceptance criteria
• status: draft, appropriate effort estimate, created and updated today
• A numeric prefix to establish order (e.g. 01-, 02-)

Decisions — one decisions/<id>.md per decision surfaced in Step 2. No confidence: inferred needed — these came directly from the session.

Step 6 — Copy selected shared decisions

For each shared decision selected in Step 3:

1. Read it from ~/.curaye/shared/decisions/
2. Copy to .curaye/decisions/<id>.md
3. Add source: shared/decisions/<id> to its frontmatter

Step 7 — Register the project

curaye link <path>

Step 8 — Report

Bootstrap complete.

  .curaye/ created at: <path>
  prd.md, stack.md, product.md: written
  Planned specs: <N> (<list of ids and titles>)
  Decisions recorded: <N> (<list>) 
  Shared decisions seeded: <list or "none">
  Registered as: <project-id>

"Run /curaye-build <first-spec-id> to start building, or /curaye-brief to review the full picture first."