setup-docs（设置文档）

One-shot bootstrap that deeply analyzes the codebase and populates all .chalk/docs/ PROFILE stubs with real, project-specific content. Also enriches chalk.json with any fields that setup-chalk couldn't auto-detect.

When to Use

• After running /setup-chalk or npx create-chalk to generate stub docs
• When PROFILE docs contain <!-- STUB --> markers or placeholder content
• When chalk.json has missing fields (routes, sourceLayout, etc.)
• To refresh docs after significant codebase changes

Workflow

Step 1: Read existing state

1. Read chalk.json — Check which fields are populated vs missing. Note any that need enrichment (especially routes, sourceLayout, dev).
2. Read each PROFILE doc — Skip any that are already populated (no <!-- STUB --> comment and more than 20 lines of real content). Note which verticals need work.
3. Read the codebase — package.json, README.md, src/ directory structure, config files, and any existing docs.

Step 2: Enrich chalk.json

Fill any missing fields by deep analysis:

• routes — If empty, scan the codebase:

- File-based routing: glob for app/**/page.{tsx,jsx} or pages/**/*.{tsx,vue} - React Router: grep for <Route path= or createBrowserRouter - Vue/Angular: grep router configs - Express/API: grep for app.get(, router. - For each route found, identify the source file/directory

• sourceLayout — If empty, scan for standard directories (src/, app/, lib/, components/, pages/, tests/)
• dev — If missing port/url, check vite.config, next.config, or framework defaults
• test — If missing, check for jest.config, vitest.config, pytest.ini, or test scripts in package.json

Write the updated chalk.json with any new fields.

Step 3: Populate Product Profile (.chalk/docs/product/PROFILE.md)

Deep analysis to populate:

• Summary: product name, one-liner, primary users, core JTBD, value prop
• Problem: what pain this solves (infer from README, package.json description)
• Target Users: table (Persona, Job, How This Helps)
• Core Jobs To Be Done: numbered list (infer from features, routes, UI patterns)
• Current Status: table of features (Feature, Status, Notes)
• What It Is / What It Is Not: clarify scope

Step 4: Populate Engineering Profile (.chalk/docs/engineering/PROFILE.md)

Deep analysis to populate:

• Architecture: process model diagram, execution contexts table, boot sequence
• Directory Structure: annotated tree with purpose per directory
• Data Flow: state management, API/IPC boundaries, storage layer
• Tech Stack: full dependency tables (runtime + dev/build), grouped by category
• Key Patterns: design patterns, error handling, testing approach

This is a single comprehensive doc — do NOT create separate architecture or techstack files.

Step 5: Populate Coding Style (.chalk/docs/engineering/coding-style.md)

Analyze 5+ representative files across different layers to document:

• File/folder naming conventions with real examples
• Component/module structure with a full code example from the codebase
• Naming conventions (variables, functions, types, files)
• Import ordering with a real example
• Export patterns
• TypeScript/language-specific patterns
• Styling approach
• Error handling patterns

Step 6: Populate AI Profile (.chalk/docs/ai/PROFILE.md)

Create the agent orientation doc:

• Project Identity: 1 paragraph summary
• Where Things Live: table (What, Where, Notes) — map every major concern to its file/directory
• Conventions to Follow: top 5-10 rules for writing code in this project
• Gotchas: numbered list of things that will surprise an agent
• How to Add a Feature: step-by-step guide based on codebase patterns

Step 7: Populate Design Profile (.chalk/docs/design/PROFILE.md)

Extract visual language:

• Color Palette: scan CSS files, Tailwind config, and components for hex codes and Tailwind classes. Organize into Primary, Neutral, and Semantic tables.
• Typography: font families, size scale, weight scale from CSS/config
• Spacing: recurring padding/margin values
• Borders & Shadows: border widths, radius, shadow definitions
• Icons: icon library and common icons used
• Component Patterns: common UI patterns (buttons, cards, panels)

Step 8: Update AGENTS.md

If AGENTS.md exists, enrich with project-specific pointers to the populated docs and critical conventions. If it doesn't exist, create a minimal one pointing to .chalk/docs/.

Step 9: Confirm

List what was populated, what was skipped, and any remaining gaps. Suggest next steps (e.g., "Design profile has placeholder content — run /update-doc design/PROFILE.md to refine").

Rules

• Write substantive content, not stubs — Every section should have real information from the codebase analysis.
• Remove <!-- STUB --> markers — When populating a doc, remove the stub comment.
• Preserve existing content — If a doc already has real content in some sections, enhance rather than overwrite.
• Use the vertical's tone — Product docs are business-facing. Engineering docs are technical. AI docs are agent-facing reference.
• Include the "Last updated" line — Format: Last updated: YYYY-MM-DD (populated from codebase analysis).
• chalk.json is authoritative — If chalk.json disagrees with a PROFILE doc, chalk.json wins. Update the doc to match.
• Real examples over abstractions — Always use actual code, file paths, and values from the codebase.

Differences from related skills

• /setup-chalk creates the .chalk/ scaffold with stubs — run it first
• /setup-docs populates those stubs with real content — run it second
• /create-doc creates a single new doc for a specific topic on demand
• /update-doc updates an existing doc with new information
• /validate-chalk checks if chalk.json and docs are complete and correct