Bootstrapper

Q4: Internationalization

• No i18n needed (single language)
• Yes — specify base locale and supported languages

If user selects yes, ask for: base locale (e.g., en) and list of supported locales (e.g., en, de, es, fr). These go in the project-specific section of CLAUDE.md, not in any atom.

Q5: Project name

Q6: Claude Code hooks (multi-select)

• Auto-format on every file edit (Recommended — PostToolUse hook)
• Auto-run tests when Claude finishes (Recommended for moderate+ — Stop hook)
• Inject git status + TODOs on session start (Recommended — SessionStart hook)
• No hooks — I'll manage manually

Q7: Subagents

• Flat team — generate planner, reviewer, and test-writer agents (Recommended for complex)
• Hierarchical team — add orchestrator agent that coordinates specialists (Recommended for enterprise/distributed)
• No — I'll work with a single Claude session

Mode 2: Spec Upload

When the user provides a spec/PRD instead of answering questions, analyze it to infer:

• Stack: from mentions of frameworks, languages, package managers
• Complexity: from entity count, integration points, business rule density
• Architecture needs: map complexity to principle recommendations
• Testing needs: from quality/compliance requirements
• i18n needs: from mentions of locales, translations, multi-language support
• Security needs: from auth mentions, user data handling, payment processing
• API design needs: from API endpoints, REST/GraphQL mentions, integrations
• Observability needs: from logging, monitoring, alerting, production deployment mentions
• Hooks needs: always recommend auto-format + auto-test for moderate+
• Subagent needs: recommend for complex+ projects

Explain each inference briefly before generating. Let the user override any inference.

Generation Process

After gathering inputs (interview or spec), generate all files in the project root.

Step 1: Compose CLAUDE.md

Read the atom files from ${CLAUDE_PLUGIN_ROOT}/skills/bootstrapper/references/atoms/ and compose:

1. Start with header: # CLAUDE.md — {project_name}
2. Add date stamp: > Auto-composed on {date}. Audit monthly. Every line must earn its place.
3. Append base/CLAUDE.md (always)
4. Append selected stack file from stacks/ with --- separator
5. Append each selected principle from principles/ with --- separators
6. If i18n was selected, add an ## i18n section with base locale, supported locales, and tooling notes (e.g., Slang for Flutter, i18next for TS, gettext for Python)
7. Add project-specific placeholder section at bottom

Write to CLAUDE.md in project root.

Token budget: Total must be under 2,500 tokens. Report word count and estimated tokens after generation.

Step 1b: Generate backbone.yml

Read the appropriate backbone template from ${CLAUDE_PLUGIN_ROOT}/skills/bootstrapper/references/backbone/{stack}.yml (use generic.yml for "None / Other" stack).

1. Replace {project_name} with the actual project name
2. If the project has known directories (from spec upload or existing codebase), update the structure section to match reality
3. Write to backbone.yml in project root

The base atom already references backbone.yml in its Context Hygiene section, so no additional CLAUDE.md wiring is needed.

Step 2: Generate Claude Code Hooks

If the user selected any hooks in Q6, read the appropriate hooks template from ${CLAUDE_PLUGIN_ROOT}/skills/bootstrapper/references/hooks/{stack}.json and generate .claude/settings.json.

Only include the hook events the user selected:

• Auto-format → PostToolUse with Write|Edit matcher
• Auto-test → Stop with build + test commands
• Session context → SessionStart with git status + TODO grep

If the user selected "None / Other" stack, use generic commands:

• Format: skip (no default formatter)
• Test: echo 'No test command configured — update .claude/settings.json'
• Session: git status only

Step 3: Generate Subagents

If the user selected subagents in Q7, read the agent templates from ${CLAUDE_PLUGIN_ROOT}/skills/bootstrapper/references/agents/ and generate:

Flat team (always when subagents selected):

• .claude/agents/planner.md — architect agent, read-only, produces implementation plans
• .claude/agents/reviewer.md — review agent, read-only, checks against CLAUDE.md
• .claude/agents/test-writer.md — test agent, can write test files only

Hierarchical team (when "Hierarchical team" selected):

• All flat team agents, plus:
• .claude/agents/orchestrator.md — coordinator agent that decomposes complex tasks, delegates to planner/reviewer/test-writer, aggregates results, and handles escalation

Adapt allowed_tools based on stack (e.g., Flutter agents get Bash(fvm:*), TS agents get Bash(npm:*), Bash(npx:*)).

Step 4: Scaffold Slash Commands

Generate .claude/commands/ with markdown files that include YAML frontmatter (description, allowed-tools).

Read command templates from ${CLAUDE_PLUGIN_ROOT}/commands/ and adapt them.

Universal commands (always include):

• commit.md — conventional commit generator
• pr.md — PR description generator

Principle-based commands (include based on selections):

• If Security atom is active: security-scan.md
• If any atom is active: review.md

Stack-specific commands (include based on stack selection):

• Flutter: include analyze.md (runs fvm flutter analyze)
• TypeScript: include typecheck.md (runs npx tsc --noEmit)
• Python: include lint.md (runs ruff check .)

Step 5: Set Up PRP Infrastructure

Create the PRP workflow scaffolding:

1. Copy INITIAL.md template from ${CLAUDE_PLUGIN_ROOT}/skills/bootstrapper/references/prp/INITIAL.md.template to project root, replacing {project_name} with the actual project name
2. Create PRPs/ directory with a .gitkeep file
3. Ensure generate-prp.md and execute-prp.md commands exist in .claude/commands/

Step 6: Create Examples Directory

Create examples/ directory with a README.md:

# Examples

Add concrete code examples here for patterns that CLAUDE.md describes.
Claude learns better from examples than from rules alone.

## Usage

Reference examples in CLAUDE.md like:
> See `examples/error-handling.ts` for the pattern.

## Structure

- One file per pattern
- Name files descriptively: `{pattern-name}.{ext}`
- Keep examples minimal — show the pattern, nothing else

Step 6b: Scaffold Skills Directory

Create .claude/skills/ with a template for the user's first skill:

1. Create .claude/skills/README.md:

# Skills

Skills are installable expertise that Claude loads on demand. If you repeat a structured prompt more than twice, turn it into a Skill.

## Creating a Skill

1. Create a directory: `.claude/skills/{skill-name}/`
2. Add a `SKILL.md` file with frontmatter (name, description) and instructions
3. Claude auto-discovers skills by matching the description to user requests

## Template

Copy `_template/SKILL.md` and customize.

2. Create .claude/skills/_template/SKILL.md:

---