bootstrap

Pra engines sem skill loading (Cursor, Codex sem MCP): leia este file inteiro e siga as instrucoes como se fossem suas. Voce nao precisa "invocar" — apenas obedeca. Cursor e Codex que tem xp-stack instalado via npm leem este SKILL.md em .cursor/rules/ ou .codex/skills/.

Voce eh o Bootstrap. Sua missao eh provisionar o stack XP/Akita neste projeto sem sobrescrever nada que ja exista. Trabalhe sempre com cp -n (no-clobber), pergunte antes de tocar arquivos pre-existentes, e reporte exatamente o que foi criado vs preservado.

XP Stack Bootstrap

Scaffold a project with the XP stack. Never overwrites existing files silently.

Environment detection (preprocessed)

Current directory: !pwd

CLAUDE.md status: !test -f "$(pwd)/CLAUDE.md" && echo "EXISTS" || echo "ABSENT"

Existing structure: !ls -la "$(pwd)/.claude" 2>/dev/null || echo "No .claude directory" !ls -la "$(pwd)/docs" 2>/dev/null || echo "No docs directory"

Your task as the agent running this skill

Follow these steps IN ORDER. Don't skip ahead.

Step 1: Collect project info

Use AskUserQuestion with these 3 questions (in one batched call):

1. Question: "What is the project name?" (the user will likely answer via Other with a kebab-case name)

   • header: "Name"
   • options: "my-api", "acme-web", "data-pipeline" — plus the automatic Other
   • multiSelect: false

2. Question: "What is the primary tech stack?"

   • header: "Stack"
   • options: "React + TypeScript + Vite", "Python + FastAPI", "Node.js + Express", "Go" — plus Other
   • multiSelect: false

3. Question: "Describe the project in one sentence."

   • header: "Description"
   • options: "General-purpose web app", "API service", "CLI tool", "Library/SDK" — plus Other for custom text
   • multiSelect: false

Store the answers as $project_name, $project_stack, $project_description (conceptually — use the values in the bash call below).

Step 2: Handle existing CLAUDE.md

Look at the CLAUDE.md status from the environment detection above.

If ABSENT: set $claude_md_action = "create" and skip to Step 3.

If EXISTS: Ask via AskUserQuestion:

• Question: "A CLAUDE.md already exists in this directory. What should we do?"
• header: "Existing"
• options:
  • "Keep existing and skip CLAUDE.md creation (Recommended)" -> skip
  • "Rename existing to CLAUDE.md.bak and create new" -> backup
  • "Abort the bootstrap entirely" -> abort
• multiSelect: false

Store the chosen action as $claude_md_action.

Step 3: Run the scaffold script

Call via Bash tool (replace the variables with the collected values):

bash ${CLAUDE_SKILL_DIR}/scripts/scaffold.sh "$(pwd)" "PROJECT_NAME_HERE" "PROJECT_STACK_HERE" "PROJECT_DESCRIPTION_HERE" "CLAUDE_MD_ACTION_HERE"

The script handles the 4 CLAUDE.md actions (create, skip, backup, abort), copies all templates with cp -n (never overwrites existing), and prints a summary of what was done.

Step 4: Report to the user

Summarize what was created / skipped. Tell them:

1. Read the new CLAUDE.md — it's the canonical source of truth for the project.
2. AGENTS.md is a symlink to CLAUDE.md (so Antigravity/Codex/Cursor read the same file). Don't edit AGENTS.md directly — edit CLAUDE.md and the symlink propagates.
3. .gitignore got 3 reserved entries (local/, .claude/wave-runs/, scripts/orchestrate/) used by the optional skills paperclip-orchestrator and local-waves — pre-existing entries are preserved.
4. Fill in the stack-specific sections (tech stack details, env vars, directory structure, lessons learned).
5. Start their first task with the task-decomposition skill.

If they want multi-agent dispatch later, mention the optional skills /xp-stack:paperclip-setup (remote async, droplet, GitHub-driven auto-merge gate) and /xp-stack:local-waves-setup (local sync, worktrees + claude -p headless).

What the scaffold script does

• Creates CLAUDE.md from the template, substituting {{PROJECT_NAME}}, {{PROJECT_STACK}}, {{PROJECT_DESCRIPTION}} with the collected values
• Copies templates/docs-tasks-template/ to docs/tasks/_template/ (recursive, no-clobber)
• Copies templates/docs-pesquisas-template/ to docs/pesquisas/_template/ (recursive, no-clobber)
• Creates .claude/settings.json from the template (no-clobber)
• Creates AGENTS.md -> CLAUDE.md symlink (and AGENTS.local.md -> CLAUDE.local.md if CLAUDE.local.md exists). Skipped when 6th arg passed as "no-symlink".
• Appends local/, .claude/wave-runs/, scripts/orchestrate/ to .gitignore if not already present (idempotent, preserves existing rules).
• Uses cp -n / cp -rn throughout — never overwrites existing files

Idempotency guarantee

Running this skill multiple times in the same directory does not modify files created in previous runs. The script always uses no-clobber copies. If the user wants to reset, they delete the files manually first.

Limits

• Only writes under $(pwd) — never touches ~/.claude/ global, never touches other repos.
• Never modifies $(pwd)/.git — the scaffold only creates source files, git is untouched.
• Never installs dependencies — if the stack needs npm install or equivalent, the user runs it after.