Skip to content

ROADMAP.md

Latest commit

 

History

History
112 lines (72 loc) · 4.44 KB

ROADMAP.md

File metadata and controls

112 lines (72 loc) · 4.44 KB

MeiGen Optimization Roadmap

Based on comparison with ui-ux-pro-max-skill — 2026-02-10

High Priority

1. README Documentation Enhancement

Current state: README covers basic usage but lacks visual examples and comparison guides.

Improvements:

  • Add real generation result showcase (before/after prompt enhancement)
  • Add Provider comparison table (MeiGen vs ComfyUI vs OpenAI pros/cons)
  • Add more usage scenario examples (logo design, product photography, batch generation)
  • Add ASCII/Mermaid workflow diagrams (generation flow, reference image flow)
  • Add tool capability overview table with statistics (7 tools, 1300+ gallery entries, 3 providers)

Reference: ui-ux-pro-max has 23KB README with feature tables, ASCII diagrams, real output examples, and anti-pattern lists.

Effort: Small | Impact: High — first impression for new users

2. Search Relevance Optimization

Current state: search_gallery uses simple string matching on 1300+ prompts in trending-prompts.json.

Improvements:

  • Implement BM25 or TF-IDF ranking for search results
  • Add category-based indexing for faster filtered searches
  • Consider splitting trending-prompts.json by category for finer-grained search
  • Add search result scoring/relevance indicator in tool response

Reference: ui-ux-pro-max uses BM25 ranking (core.py, 253 lines) with regex domain detection and autocomplete. Zero external dependencies.

Effort: Medium | Impact: High — directly improves gallery discovery experience

Medium Priority

3. Multi-Platform MCP Setup — DONE

Implemented: npx meigen init <platform> — integrated into existing package, zero new dependencies.

Supported platforms:

  • Cursor — .cursor/mcp.json (mcpServers format)
  • VS Code / GitHub Copilot — .vscode/mcp.json (servers format + type: "stdio")
  • Windsurf — ~/.codeium/windsurf/mcp_config.json (global, mcpServers format)
  • Roo Code — .roo/mcp.json (mcpServers format)
  • Claude Code — .mcp.json (mcpServers format)

Key design decisions:

  • No separate CLI package — subcommand of existing meigen binary (~80 lines)
  • Merges into existing config files without overwriting other servers
  • All platforms use the same npx -y meigen@latest server definition, only wrapper format and file path differ

Files: src/cli/init.ts, bin/meigen-mcp.js

Effort: Small (was estimated Large) | Impact: High — expands user base beyond Claude Code

Low Priority

4. Brand Style Persistence (Instructions Update)

Current state: Each generation is independent; no mechanism to maintain style consistency across a multi-image project.

Improvements:

  • Add "Brand Style Master" workflow guidance in SERVER_INSTRUCTIONS
  • Guide LLM to create a style reference file (colors, typography, mood, composition rules)
  • Subsequent generations auto-reference the style file for consistency
  • No code changes needed — pure instruction/skill update

Reference: ui-ux-pro-max uses Master + Overrides pattern (MASTER.md + pages/*.md) for hierarchical design system persistence.

Effort: Small | Impact: Medium — valuable for professional multi-image workflows

5. Skill Activation Strategy

Current state: visual-creative skill triggers on image-related keywords.

Improvements:

  • Review marketplace.json for strict mode configuration
  • Test activation reliability across edge-case prompts
  • Expand trigger keywords if needed (e.g., brand design, mockup, visual identity)
  • Consider adding strict: false if not already set

Reference: ui-ux-pro-max uses "strict": false in marketplace.json for broader skill activation.

Effort: Small | Impact: Low — marginal improvement

Not Planned

Item Reason
Python search engine Already have TypeScript MCP server, no need for Python
Symlink strategy npx distribution eliminates duplication problem
CSV data format JSON is better for our prompt library (nested data, image URLs)

Notes

  • Priorities may shift based on user feedback and usage data
  • "Effort" is relative: Small (~1 day), Medium (~3 days), Large (~1 week+)
  • All changes should follow the project's "Instructions over Code" philosophy — prefer updating SKILL.md / SERVER_INSTRUCTIONS over adding code branches