Lotus

_{Art: Purity card from Slay the Spire 2 — "Remove a card from your deck."}

繁體中文

Minimal AI config recommender for coding assistants.

The AI coding tool ecosystem is noisy — thousands of skills, dozens of MCP registries, new frameworks weekly. Lotus answers one question: given your project, what's the minimum that works?

What It Does

1. Analyzes your project — detects language, framework, database, CI, and existing AI config
2. Recommends the minimum effective set of skills, agents, MCP servers, and hooks from a curated catalog
3. Compares competing solutions by token cost, time, and quality
4. Applies configs directly to your .claude/ directory

Install

go install github.com/texliao/lotus/cmd/lotus@latest

Usage

# Detect stack and scan existing AI config
lotus analyze .

# Get recommendations
lotus recommend .

# Preview changes without writing
lotus apply . --dry-run

# Apply recommended config
lotus apply .

# Browse catalog
lotus catalog list
lotus catalog list --kind bundle
lotus catalog list --stack go

# Show entry details
lotus catalog show superpowers

Example

$ lotus analyze .

Project: /home/user/my-go-api

Detected stacks:
  LANGUAGE  VERSION  FRAMEWORK  DATABASE  CI
  go        1.24     gin        postgres  github-actions

Existing AI config:
  CLAUDE.md:    true
  Skills:       1 (git-commit)
  Agents:       none
  Hooks:        none
  MCP servers:  none

Inferred use cases: backend-development, code-review, testing, git-workflow

$ lotus recommend .

Recommendations for /home/user/my-go-api
Detected: go

  SCORE  ACTION  KIND    ID           REASON
  -----  ------  ----    --           ------
  22     add     bundle  superpowers  matches use case: code-review; stack-agnostic

1 recommendations. Run `lotus apply .` to apply.

Catalog

Lotus ships with a curated catalog of evaluated AI coding tools:

Kind	Description	Examples
skill	Single SKILL.md file	minimax-frontend-dev, git-commit
bundle	Multi-file package (skills + agents + hooks)	superpowers, gstack, D-Team
source	Large library to cherry-pick from	agency-agents (144 personas)
agent	Single agent definition	-
mcp-server	MCP server config	-
hook	Shell hook for Claude Code	-

Included Entries

Bundles

• superpowers — structured dev workflow (brainstorm, plan, TDD, review). Zero deps.
• gstack — virtual eng team (plan, build, QA, ship, retro). Requires Bun + Playwright.
• A-Team — meta-agent that generates custom agent teams via interview.

Skills

• MiniMax-AI/skills — frontend, fullstack, Android, iOS, Flutter, React Native, shader, PDF/PPTX/XLSX/DOCX generation.
• git-commit — conventional commit workflow.

Sources

• agency-agents — 144 agent personas across engineering, design, marketing, sales, QA, game dev, and more.

How Recommendations Work

1. Stack detection — scan go.mod, package.json, Cargo.toml, pyproject.toml, CI configs, Docker Compose
2. Use case inference — map detected stack to use cases (backend-development, frontend-development, etc.)
3. Catalog matching — find entries matching your use cases and stack
4. Scoring — base_score = use_case_match + stack_match + tier_bonus - weight_penalty
5. Conflict resolution — if two entries conflict (e.g., superpowers vs gstack), keep the higher-scored one

Adding Entries

Add a YAML file to catalogdata/data/<kind>/:

id: my-skill
kind: skill
name: "My Skill"
source:
  registry: github
  repo: "user/repo"
  url: "https://github.com/user/repo"
use_cases:
  - backend-development
stacks:
  - go
requires:
  tools: []
  mcp_servers: []
  runtime: []
lotus:
  tier: recommended
  notes: "Description of what this does and why."
  conflicts_with: []
  pairs_well_with: []

License

MIT