manifest-automation

Automate the creation and updating of manifest files for AI coding tools using advanced browser automation, web search, and intelligent field-level retry logic.

Overview

This skill extends beyond the existing manifest-creator by adding:

• Dual Mode Operation: CREATE new manifests or UPDATE existing ones with smart merge
• Advanced Web Automation: Playwright MCP for dynamic content, pricing calculators, and marketplace navigation
• Intelligent Retry Logic: 3-attempt strategy per field with graceful degradation to TODO comments
• Smart Merge: UPDATE mode preserves user-curated data while refreshing auto-discoverable fields
• Type-Optimized Workflows: Dedicated extraction strategies for each manifest type

Supported Manifest Types

1. CLI - Command-line tools with platform-specific install commands
2. Extension - IDE extensions with marketplace URLs and install URIs
3. IDE - Integrated development environments with GUI installers
4. Model - AI models with technical specs and token pricing
5. Provider - Model providers (foundation or service)
6. Vendor - Companies and organizations

Usage

CREATE Mode

Create a new manifest from scratch:

# CLI tool
node .claude/skills/manifest-automation/scripts/automate.mjs create cli cursor-cli https://cursor.com/cli

# Extension
node .claude/skills/manifest-automation/scripts/automate.mjs create extension copilot https://github.com/features/copilot

# Model
node .claude/skills/manifest-automation/scripts/automate.mjs create model claude-opus https://anthropic.com/claude

# Provider
node .claude/skills/manifest-automation/scripts/automate.mjs create provider openrouter https://openrouter.ai

# IDE
node .claude/skills/manifest-automation/scripts/automate.mjs create ide cursor https://cursor.com

# Vendor
node .claude/skills/manifest-automation/scripts/automate.mjs create vendor anthropic https://anthropic.com

UPDATE Mode

Update an existing manifest with fresh data:

# Update existing CLI manifest
node .claude/skills/manifest-automation/scripts/automate.mjs update cli cursor-cli https://cursor.com/cli

# Update extension (URL optional if already in manifest)
node .claude/skills/manifest-automation/scripts/automate.mjs update extension copilot

How It Works

CREATE Mode Workflow

1. Load Schema & Workflow

   • Read JSON schema for manifest type
   • Load type-specific workflow instructions
   • Initialize field tracker for retry logic

2. Extract Information

   • Use Playwright MCP to navigate websites
   • Use WebSearch to discover GitHub repos, marketplaces, social media
   • Extract data following type-specific workflow
   • Track attempts per field (max 3)

3. Handle Failures Gracefully

   • After 3 failed attempts per field: add TODO comment
   • Continue with partial data (save as draft)
   • Generate completion report

4. Validate & Save

   • Write manifest JSON to appropriate path
   • For IDE/CLI/Extension: check vendor manifest exists, create if missing
   • Update data/github-stars.json
   • Run pnpm generate to regenerate TypeScript data files
   • Run schema validation
   • Report success/failures to user

UPDATE Mode Workflow

1. Load Existing Manifest

   • Read current manifest from disk
   • Parse and validate structure

2. Extract Fresh Data

   • Follow same extraction workflow as CREATE
   • Use Playwright and WebSearch for current data

3. Smart Merge

   • AUTO_UPDATE: Replace with new values (versions, descriptions, pricing, specs)
   • PRESERVE: Keep existing (id, name, verified, i18n, relatedProducts)
   • MERGE_ADDITIVE: Add new items to arrays/objects (communityUrls, platforms)
   • CONDITIONAL: Present both for manual review (license)

4. Generate Change Report

   • Show what was updated, added, preserved
   • Flag fields needing manual review
   • Validate and save merged manifest

Field Extraction Strategies

CLI Workflow Focus

• Platform-specific install commands (brew, npm, apt, winget)
• Launch commands and installation paths
• GitHub releases for version tracking
• Cross-platform support detection

Extension Workflow Focus

• IDE marketplace URLs (VS Code, JetBrains, Open VSX)
• Install URIs (vscode:extension/, cursor:extension/)
• Supported IDE detection and compatibility
• Publisher and extension ID extraction

Model Workflow Focus

• Technical specifications (parameter size, context window, max output)
• Token pricing ($/M tokens for input/output/cache)
• Platform URLs (HuggingFace, Artificial Analysis, OpenRouter)
• Model card information extraction

Provider Workflow Focus

• Provider type detection (foundation vs service)
• API key application URLs
• Platform integration discovery
• Developer documentation

IDE Workflow Focus

• GUI installation (DMG, EXE, DEB installers)
• Download page navigation
• App Store distributions
• GUI-based installation steps

Vendor Workflow Focus

• Company mission and description
• Official social media accounts
• Community links and presence
• Organization-level information

Tool Integration

When to Use Each Tool

Playwright MCP (mcp__playwright__browser_*):

• Dynamic JavaScript-rendered content
• Multi-step navigation (tabs, dropdowns, modals)
• IDE marketplace pages
• Pricing calculators and forms
• Interactive website elements
• LinkedIn company pages (verification and metadata)

WebSearch:

• Discovering GitHub repositories
• Finding marketplace URLs
• Locating social media profiles
• Platform-specific pages (HuggingFace, npm, PyPI)
• Validation and alternatives
• Wikipedia for cross-referencing company information
• LinkedIn company pages for authoritative metadata

WebFetch (fallback):

• Static documentation pages
• Simple content fetching
• Direct URL access for known pages

Reference-Only Sources

Wikipedia:

• Use for verifying company descriptions and basic facts
• Cross-reference founding dates, headquarters, official names
• Validate social media links listed in Wikipedia infobox
• NEVER store Wikipedia URLs in manifests - use only for verification

LinkedIn Company Pages:

• Primary authoritative source for company metadata
• Verify company size, industry, official description
• Store the LinkedIn URL in communityUrls.linkedin
• Use company description to validate website description accuracy

Playwright Patterns

Pattern 1: Navigate → Snapshot → Extract

1. mcp__playwright__browser_navigate(url)
2. mcp__playwright__browser_snapshot()
3. Extract data from snapshot elements

Pattern 2: Interactive Navigation

1. Take initial snapshot
2. mcp__playwright__browser_click(element)
3. mcp__playwright__browser_wait_for(text)
4. Take new snapshot and extract

Pattern 3: Form Interaction

1. mcp__playwright__browser_fill_form(fields)
2. Click submit button
3. Extract results from response

Retry Logic

3-Attempt Strategy

Each field gets up to 3 extraction attempts before falling back to TODO comment:

Attempt 1: Direct extraction from primary source

• Official website, documentation, or main page
• Most authoritative source

Attempt 2: Alternative sources or search

• WebSearch for the information
• Alternative URL patterns
• Related pages or sections

Attempt 3: Final fallback sources

• Third-party platforms
• Package manager websites
• Community resources

After 3 failures: Add TODO comment with reason

Retry Examples

Installation Commands:

1. Official installation docs page
2. WebSearch for package manager (brew, npm, PyPI)
3. GitHub README or package manager sites
4. TODO if still not found

Pricing:

1. Official pricing page
2. Footer/header navigation links
3. Documentation pricing mentions
4. TODO if behind "Contact Sales"

Community URLs:

1. Website footer social links
2. WebSearch "<product> official <platform>"
3. Common URL patterns validation
4. Set to null if not found

TODO Comment Format

{
  "communityUrls": {
    "discord": null, // TODO: Could not auto-discover after 3 attempts. Not found in footer or search results.
    "twitter": "https://x.com/product"
  }
}

Smart Merge Field Categories

AUTO_UPDATE Fields

Always replaced with freshly discovered values:

• latestVersion - Always fetch latest
• description - From official source
• websiteUrl - Official URL
• docsUrl - Documentation URL
• tokenPricing - Model pricing (changes frequently)
• size, contextWindow, maxOutput - Model specs

PRESERVE Fields

Never updated (user-curated):

• id - Core identifier
• name - Display name
• verified - Manual verification status
• i18n - User translations
• relatedProducts - Manually curated relationships

MERGE_ADDITIVE Fields

Add new items, keep existing:

• communityUrls - Add newly discovered social links
• platformUrls - Add new platform integrations
• supportedIdes - Add new IDE support
• platforms - Add new OS support
• pricing - Add new tiers, preserve existing

CONDITIONAL Fields

Require manual review:

• license - Only update if from authoritative source (GitHub)
• vendor - Review name changes carefully

Error Handling

Graceful Degradation

1. Field-Level Failures: Track each field independently
2. Partial Manifests: Save draft even with missing fields
3. TODO Comments: Mark failed fields with reason
4. Completion Reports: Show what succeeded/failed

Completion Report Format

📊 Manifest Automation Report
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Status: DRAFT (3 fields incomplete)

✅ Successfully Extracted (42 fields)
❌ Failed Extraction (3 fields):
   1. communityUrls.discord (3 attempts)
      Reason: Not found in website footer or WebSearch
   2. pricing[2].value (3 attempts)
      Reason: Enterprise pricing behind contact form
   3. platforms[1].installCommand (3 attempts)
      Reason: Windows installation docs incomplete

⚠️  Cross-Reference Validation:
   ✓ Wikipedia: Company description verified
   ✓ LinkedIn: Company metadata matches
   ⚠️ Description length mismatch: Website (180 chars) vs LinkedIn (220 chars)

🌐 i18n Consistency Warnings (UPDATE mode only):
   ⚠️ English description updated - review i18n translations
   ⚠️ i18n.zh-Hans may be outdated (last updated: previous description)
   ⚠️ i18n.de may be outdated (last updated: previous description)
   → Add TODO comments to i18n fields requiring manual review

📝 Next Steps:
1. Review manifest file: manifests/clis/cursor-cli.json
2. Manually fill TODO-marked fields if information available
3. Add i18n translations (zh-Hans, de, ko)
4. Update verified field once data confirmed accurate
5. Run validation: pnpm test:validate

Note: Vendor manifest, github-stars.json, and TypeScript files already updated.

Comparison with manifest-creator

| Feature | manifest-creator | manifest-automation | |---------|------------------|---------------------| | Tools | WebFetch only | Playwright MCP + WebSearch + WebFetch | | Modes | CREATE only | CREATE + UPDATE | | Retry Logic | Manual/ask user | Automatic 3-attempt per field | | Dynamic Content | Limited | Full Playwright support | | Merge Logic | N/A | Smart field categorization | | Error Handling | Stop and ask | TODO + draft + report | | Workflows | Generic 7-step | Type-optimized phases | | Community Discovery | Basic | Advanced WebSearch | | Pricing Extraction | Static pages | Dynamic calculators | | Marketplace Support | Limited | Full automation (VS Code, JetBrains) |

Best Practices

Accuracy

• Never generate or guess data - always extract from authoritative sources
• Validate URLs exist and are official
• Verify version numbers from releases, not estimates
• Check social media accounts are official (verification badges, content)
• Use Wikipedia for cross-referencing company facts, model specs, and historical data
• Use LinkedIn as authoritative source for company metadata and descriptions

Completeness

• Follow type-specific workflow completely
• Don't skip optional fields - attempt extraction
• Use full 3-attempt retry strategy before TODO
• Generate comprehensive completion reports
• Cross-reference with Wikipedia and LinkedIn to ensure no critical information is missed
• Validate i18n translations match English default content

Validation

• Always run schema validation after manifest creation
• Fix validation errors before marking complete
• Verify enum values (os, ideId, pricing categories)
• Check URL formats (https://, proper domains)
• Verify company information against Wikipedia and LinkedIn
• Ensure i18n consistency - all languages must reflect the same content

Smart Merge

• In UPDATE mode, trust the merge categories
• Review CONDITIONAL fields manually
• Don't blindly replace PRESERVE fields
• Validate MERGE_ADDITIVE doesn't create duplicates
• Flag i18n fields for review when English content changes
• Include i18n drift warnings in completion reports

Cross-Referencing Strategy

1. Primary source: Official website and documentation
2. Verification source 1: LinkedIn company page (for vendors/companies)
3. Verification source 2: Wikipedia (for well-known entities)
4. Conflict resolution: Official website > LinkedIn > Wikipedia
5. Never store reference URLs: Wikipedia and verification sources are for validation only

Troubleshooting

Playwright Navigation Fails

• Wait 2 seconds, retry navigation
• Check for redirects or authentication walls
• Try alternative URLs or WebSearch fallback

Element Not Found in Snapshot

• Take screenshot for debugging
• Search page text directly (not by element ref)
• Try WebSearch as alternative approach

Dynamic Content Doesn't Load

• Use mcp__playwright__browser_wait_for with specific text
• Scroll page to trigger lazy loading
• Click "Load More" or "Show All" buttons
• Fallback to WebSearch

Rate Limiting or Blocking

• Add delays between requests
• Use WebSearch as alternative data source
• Mark field as TODO if consistently blocked

Files Created

After running this skill, expect:

1. Manifest file: manifests/<type>s/<name>.json
2. With TODO comments: For failed field extractions
3. Valid against schema: manifests/$schemas/<type>.schema.json
4. Ready for validation: Run pnpm test:validate

GitHub Stars Update

After creating or updating a manifest, you MUST update the data/github-stars.json file:

Why Update github-stars.json?

The github-stars.json file tracks GitHub star counts for all manifests. When you create or update a manifest:

• A new entry must be added for tracking
• The entry is initialized with null (stars not yet fetched)
• A scheduled job will automatically fetch the actual star count later

How to Update

Use the updateGithubStarsEntry() function from github-stars-updater.mjs:

import { updateGithubStarsEntry } from './lib/github-stars-updater.mjs'

// After creating a new manifest
const result = updateGithubStarsEntry('cli', 'cursor-cli', { isNew: true })

// After updating an existing manifest
const result = updateGithubStarsEntry('extension', 'claude-code', { isNew: false })

CLI Usage

You can also use the CLI directly:

# Add new entry
node .claude/skills/manifest-automation/scripts/lib/github-stars-updater.mjs add cli cursor-cli

# Update existing entry
node .claude/skills/manifest-automation/scripts/lib/github-stars-updater.mjs update extension claude-code

# Remove entry
node .claude/skills/manifest-automation/scripts/lib/github-stars-updater.mjs remove ide windsurf

What It Does

The updater will:

1. Load the current github-stars.json file
2. Add or update the entry: <category>["<id>"] = null
3. Sort entries alphabetically within the category
4. Save the updated file back to disk

Example Result

For a new CLI manifest cursor-cli:

{
  "clis": {
    "claude-code-cli": 43.5,
    "cursor-cli": null,  // ← newly added
    "kode": 3.6
  }
}

Integration with Workflow

The automate.mjs script exports the necessary information for you to call the updater:

// After workflow completes and manifest is saved:
import {
  updateGithubStarsEntry,
  manifestType,
  manifestName,
  operationMode
} from './automate.mjs'

updateGithubStarsEntry(manifestType, manifestName, { isNew: operationMode === 'create' })

Next Steps After Creation

1. Review manifest: Check all extracted values for accuracy
2. Fill TODOs: Manually add fields that couldn't be auto-discovered
3. Add translations: Populate i18n object with localized content (zh-Hans, de, ko)
4. Set verified: Change verified to true if data is confirmed accurate
5. Add related products: Manually curate relatedProducts array
6. Run validation: Ensure schema compliance with pnpm test:validate
7. Commit changes: Add manifest to git repository

Note: Vendor manifest and github-stars.json are automatically handled during creation.

i18n Consistency Requirements

IMPORTANT: When creating or updating manifests with i18n content:

1. Default Language (English):

   • All English content should be in top-level fields (e.g., description, name)
   • English should NEVER appear in the i18n object

2. Multi-Language Support:

   • Support at least 3 languages: English (default), Simplified Chinese (zh-Hans), German (de)
   • NEVER hardcode only 'en' | 'zh-Hans' - always include de and support extensibility

3. Content Synchronization:

   • All i18n translations must reflect the SAME content as the English default
   • When updating default English fields, flag i18n fields for manual review
   • Never auto-translate - mark translations as TODO if not manually provided

4. Validation Checks:

   • After UPDATE mode, verify i18n content hasn't become stale
   • Flag i18n fields that may need updating when English content changes
   • Include i18n consistency warnings in completion reports

5. Structure Example:

{
  "name": "Example Product",
  "description": "This is the default English description",
  "i18n": {
    "zh-Hans": {
      "name": "示例产品",
      "description": "这是默认的英文描述"
    },
    "de": {
      "name": "Beispielprodukt",
      "description": "Dies ist die englische Standardbeschreibung"
    }
  }
}

In UPDATE mode:

• If default English fields change, add comment: // TODO: Update i18n translations to match new English content
• Report i18n drift in completion report
• Preserve existing i18n content (PRESERVE category) but flag for review