convez.ai
SkillProdutividadeIntermediário
fallow-rs

Fallow Skills

fallow-rs/fallow-skills

Inteligência de codebase para agentes JS/TS — código não usado, desvio arquitetural, complexidade em ms

Ver no GitHub
BomPluginScore · de 0 a 100

Skills para agentes de IA que ensinam como analisar bases JavaScript e TypeScript usando inteligência de codebase Fallow. Detecta código não utilizado, duplicações, dependências circulares, pontos de complexidade e desvios arquiteturais. Compatível com Claude Code, Cursor, Windsurf e 30+ agentes.

Instalação

$npx convez-ai fallow-rs/fallow-skills

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
141
Forks
11
Acompanhando
1
Saúde
50%
Último commit
há 1 dia
Versão
v1.4.41
Issues abertas
2
Contribuidores
3

Casos de uso

  • Identificar e remover código morto automaticamente
  • Detectar duplicação de código e refatorar
  • Encontrar dependências circulares e violações arquiteturais
  • Analisar riscos em código modificado
  • Priorizar hotspots de complexidade para refatoração

Veredito

Ferramenta sólida e bem integrada no ecossistema de agentes. O Fallow oferece análise estática rápida com suporte a runtime opcional; as skills educam agentes sobre como usar a ferramenta efetivamente. Recomendado para times que querem automação séria de limpeza e análise de codebase.

README

A tradução para o português deste README ainda está na fila.

Agent skills for fallow, Rust-native codebase intelligence for JavaScript and TypeScript. The free static layer reports quality, changed-code risk, cleanup opportunities, circular dependencies, code duplication, complexity hotspots, and architecture boundary violations in milliseconds. The optional paid runtime layer (Fallow Runtime) adds production execution evidence so agents can delete cold code, flag hot-path changes, and retire stale flags with proof. 118 framework plugins, zero configuration. Works with any agent that supports the Agent Skills specification: Claude Code, Cursor, OpenAI Codex, Windsurf, GitHub Copilot, Gemini CLI, Amp, and 30+ more.

Linters enforce style. Formatters enforce consistency. Fallow enforces relevance. Linters work file by file. TypeScript works type by type. Neither builds the full module graph, so neither can see what nothing depends on. Fallow does, in milliseconds. These skills teach agents how to use fallow effectively: which commands to run, what flags to use, how to interpret output, and how to avoid common pitfalls.

Quick Start

Agent Skills CLI

npx skills add fallow-rs/fallow-skills

Claude Code

/install fallow-rs/fallow-skills

Agent-specific CLI shortcuts

npx skills add fallow-rs/fallow-skills --agent windsurf
npx skills add fallow-rs/fallow-skills --agent amp
gemini skills install https://github.com/fallow-rs/fallow-skills.git

Manual install

tmp=$(mktemp -d)
git clone https://github.com/fallow-rs/fallow-skills.git "$tmp/fallow-skills"

Copy the fallow skill directory into your agent's skills folder:

# OpenAI Codex, Amp, and agents using the shared Agent Skills location
mkdir -p ~/.agents/skills
cp -R "$tmp/fallow-skills/fallow/skills/fallow" ~/.agents/skills/fallow

# Claude Code
mkdir -p ~/.claude/skills
cp -R "$tmp/fallow-skills/fallow/skills/fallow" ~/.claude/skills/fallow

# Cursor
mkdir -p ~/.cursor/skills
cp -R "$tmp/fallow-skills/fallow/skills/fallow" ~/.cursor/skills/fallow

# Windsurf
mkdir -p ~/.codeium/windsurf/skills
cp -R "$tmp/fallow-skills/fallow/skills/fallow" ~/.codeium/windsurf/skills/fallow

# GitHub Copilot
mkdir -p .github/skills
cp -R "$tmp/fallow-skills/fallow/skills/fallow" .github/skills/fallow

Use npx skills add fallow-rs/fallow-skills --all for installer-managed discovery, or copy fallow/skills/fallow into your agent's skills location as fallow. This skill follows the open Agent Skills specification and works with any compatible agent.

Prerequisites

Fallow must be installed in the target project:

npm install -g fallow    # prebuilt binaries
npx fallow                   # or run without installing

See the installation guide for all options including cargo install fallow-cli.

Available Skills

SkillDescriptionTrigger phrases
fallowCodebase intelligence for JS/TS: quality, changed-code risk, cleanup opportunities, circular deps, duplication, complexity, and (with Runtime) hot-path and cold-path evidence"check code health", "audit this PR", "find cleanup opportunities", "find duplicates", "what code actually runs"

What's Included

fallow

CategoryWhat it does
CleanupFind unused files, exports, types, dependencies, enum/class members, stale suppressions, and other safe cleanup candidates
DuplicationFind code clones with 4 modes: strict, mild, weak, semantic
ComplexityFunction complexity analysis, hotspot detection, health scores
Auto-FixRemove unused exports and dependencies with dry-run preview
CIGitHub Actions, SARIF upload, baseline comparison, PR-scoped checks
MonorepoPer-workspace analysis with cross-package resolution
DebugTrace export usage chains, file edges, and dependency usage

Reference Documentation

  • CLI Reference: all 10 commands, flags, JSON output structure, config format
  • Gotchas: 19 pitfalls with WRONG/CORRECT examples
  • Patterns: 14 workflow recipes for CI, monorepos, migration, incremental adoption

Example Prompts

Once installed, you can use natural language:

  • "Audit the codebase quality"
  • "Are there any unused dependencies?"
  • "Find code duplication in the codebase"
  • "Clean up unused exports"
  • "Set up a CI quality gate"
  • "Check the complexity of this codebase"
  • "Why is this export flagged as unused?"
  • "Check if this PR introduces quality risk"
  • "Find unused files in the payments package"
  • "What's the duplication percentage?"

How It Works

User: "Find all unused exports"
  ↓
Agent loads fallow skill
  ↓
Skill instructs: run `fallow dead-code --format json --quiet --unused-exports`
  ↓
Agent executes command, parses JSON output
  ↓
Agent summarizes findings with file paths and line numbers

The skill provides agents with:

  1. Command knowledge: which fallow command + flags to use for each task
  2. Output parsing: how to interpret JSON results
  3. Guardrails: always dry-run before fix, never run watch, use --yes in non-TTY
  4. Debugging: how to trace false positives with --trace

Contributing

See CLAUDE.md for repository structure, skill creation guidelines, and quality standards.

Related

License

MIT. See LICENSE for details.

Ver o original no GitHub

Bom para

Relacionados