fourcolors Skills
fourcolors/skills
Skills modulares e otimizadas para assistentes de IA sem poluir contexto
Ver no GitHubPacote curado de skills production-grade para assistentes de IA (Claude, Cursor, Windsurf). Inclui capacidades especializadas como gerenciamento de scratch files, orquestração multi-agente (ping-pong), memória distribuída de subagentes e geração de artefatos HTML para planejamento visual.
Instalação
npx convez-ai fourcolors/skills1 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
Casos de uso
- Injetar comportamentos especializados em assistentes de código
- Orquestrar pair programming e QC com múltiplos agentes
- Manter memória distribuída entre subagentes
- Gerar artefatos visuais de planejamento em HTML estruturado
Veredito

Projeto maduro e bem-documentado que resolve um problema real: injetar conhecimento processual em assistentes de IA sem comprometer janelas de contexto. A arquitetura de progressive disclosure é elegante e as skills fornecidas (scratch, ping-pong, subagent-memory, html-discussion) cobrem cenários genuínos de desenvolvimento colaborativo.
README
A tradução para o português deste README ainda está na fila.
███████╗ ██████╗ ██╗ ██╗██████╗ ██████╗ ██████╗ ██╗ ██████╗ ██████╗ ███████╗
██╔════╝██╔═══██╗██║ ██║██╔══██╗██╔════╝██╔═══██╗██║ ██╔═══██╗██╔══██╗██╔════╝
█████╗ ██║ ██║██║ ██║██████╔╝██║ ██║ ██║██║ ██║ ██║██████╔╝███████╗
██╔══╝ ██║ ██║██║ ██║██╔══██╗██║ ██║ ██║██║ ██║ ██║██╔══██╗╚════██║
██║ ╚██████╔╝╚██████╔╝██║ ██║╚██████╗╚██████╔╝███████╗╚██████╔╝██║ ██║███████║
╚═╝ ╚═════╝ ╚═════╝ ╚═╝ ╚═╝ ╚═════╝ ╚═════╝ ╚══════╝ ╚═════╝ ╚═╝ ╚═╝╚══════╝
███████╗██╗ ██╗██╗██╗ ██╗ ███████╗
██╔════╝██║ ██╔╝██║██║ ██║ ██╔════╝
███████╗█████╔╝ ██║██║ ██║ ███████╗
╚════██║██╔═██╗ ██║██║ ██║ ╚════██║
███████║██║ ██╗██║███████╗███████╗███████║
╚══════╝╚═╝ ╚═╝╚═╝╚══════╝╚══════╝╚══════╝
fourcolors skills
A curated pack of production-grade, highly optimized capabilities and workflow guidelines designed for AI coding assistants (such as Claude Code, Cursor, Windsurf, and others).
This repository is maintained by fourcolors and is fully installable and manageable with the open skills CLI.
How AI Agent Skills Work
Agent skills are a standardized method of injecting procedural knowledge and specialized behaviors into AI coding assistants without polluting their permanent context window. They operate under two primary principles:
1. Declarative & Procedural Instruction (SKILL.md)
Each skill is organized in a directory containing a structured SKILL.md file. While the human-facing README.md documents general features, the SKILL.md uses specialized frontmatter (YAML) and system-instruction language specifically designed to guide LLMs and autonomous agents on how to execute tasks safely, repeatably, and cleanly.
2. Progressive Disclosure
To prevent prompt bloat and keep context windows highly efficient:
- At the start of a session, the AI agent only preloads the skill’s frontmatter metadata (the
nameanddescription). - The full instruction set inside
SKILL.mdis not loaded initially. - Instead, the agent detects when a user request matches the skill's metadata and dynamically issues a
view_fileor fetch call to load the complete instructions only when relevant.
Active Skills Index
| Skill | Category | Description |
|---|---|---|
scratch | Workflow | Light, project-local .scratch/ folder convention for temporary development scripts and artifacts without polluting the main codebase. |
ping-pong | Multi-Agent | A robust pair programming and quality-control simulation orchestrating specialized navigator (pp-ping), driver (pp-pong), and QC auditor (pp-auditor) subagents. |
subagent-memory | Cognition | An observational memory discipline for subagents using date-grouped entries, priority emojis, and a CLI reflection pass to avoid context cliff issues. |
html-discussion | Visual Planning | Script-driven, byte-addressable HTML planning artifact generator supporting option comparison tables, neomorphic/glassmorphism themes, and structured snippets. |
Development & Execution Lifecycle
This diagram illustrates how all the packaged skills (html-discussion, ping-pong, subagent-memory, and scratch) and custom slash commands (/discussion, /crystallize) coordinate dynamically throughout your system development process:
graph TD
subgraph Planning ["1. Scaffolding & Planning (HTML-Discussion)"]
A["/discussion {slug}"] -->|Scaffold Draft| B["docs/discussions/{date}-{slug}.html"]
B -->|Append components via bin/ scripts| C["Add comparison tables, grids, KPIs"]
C -->|Decisions made| D["/crystallize {slug}"]
D -->|Propose spec changes| E["Update docs/architecture.html & features/"]
end
subgraph Implementation ["2. Pair Programming Loop (Ping-Pong & Subagent Memory)"]
E -->|Team Lead dispatches Scenario Trio| F["pp-ping (Navigator)"]
F -->|Discover test suite & write FAILING test| G["RED State (In-place Spec)"]
G -->|Driver takes over| H["pp-pong (Driver)"]
H -->|Implement code until tests pass| I["GREEN State (Passing Test)"]
I -->|Audit is triggered| J["pp-auditor (Sanity Check)"]
J -->|Blocking axes: On task, Correct, Right, Smart + Extra-mile advisory| K{"Audit PASS?"}
K -->|FAIL| L["Re-dispatch ping (spec gap) or pong (impl gap)"]
L --> F
K -->|PASS| M["Merge and update GOAL.md"]
F -.->|Dated bullets| Memory["subagent-memory (MEMORY.md)"]
H -.->|Dated bullets| Memory
J -.->|Dated bullets| Memory
end
subgraph Shipping ["3. Traceability & Ship"]
M -->|All scenarios completed| N["/discussion ship {slug}"]
N -->|Manifest updated to shipped| O["Stamp shipped_at & shipped_commit"]
O -->|Re-indexing| P["docs/discussions/INDEX.html Rebuilt"]
P -->|Executable provenance| Q["@scope:{slug} tags remain on features forever"]
end
classDef planning fill:#25252e,stroke:#d4a853,stroke-width:2px,color:#fff;
classDef impl fill:#1e1e26,stroke:#60a5fa,stroke-width:2px,color:#fff;
classDef ship fill:#14141a,stroke:#4ade80,stroke-width:2px,color:#fff;
class A,B,C,D,E planning;
class F,G,H,I,J,K,L,M,Memory impl;
class N,O,P,Q ship;
Repository Layout
skills/
├── LICENSE
├── README.md <-- You are here
└── skills/
├── scratch/
│ ├── SKILL.md <-- Agent-facing rules
│ └── templates/
│ └── README.md
├── ping-pong/
│ ├── SKILL.md <-- Orchestrator guidelines
│ ├── README.md <-- Human instructions
│ ├── agents/ <-- Subagent definitions
│ │ ├── pp-ping.md
│ │ ├── pp-pong.md
│ │ └── pp-auditor.md
│ └── references/ <-- Loaded on demand by the lead, not at trigger time
│ ├── briefs.md
│ ├── audit-modes.md
│ └── monitoring.md
├── subagent-memory/
│ ├── SKILL.md <-- Memory rules
│ ├── README.md <-- Human instructions
│ └── templates/
│ └── MEMORY.md
└── html-discussion/
├── SKILL.md <-- HTML artifact rules
├── README.md <-- Human instructions
├── bin/ <-- Mutation shell scripts
├── snippets/ <-- HTML slot component blueprints
├── themes/ <-- Glassmorphism & neomorphism CSS themes
└── commands/ <-- Custom slash command definitions (crystallize, discussion)
Installation
Install individual skills globally or locally inside your projects using the skills CLI:
Project-Scoped (Recommended)
Project-scoped installations commit the skill metadata locally so anyone working on the repository shares the same agent guidelines.
# Install Scratch
npx skills add fourcolors/skills --skill scratch -a claude-code
# Install Ping-Pong
npx skills add fourcolors/skills --skill ping-pong -a claude-code
# Install Subagent Memory
npx skills add fourcolors/skills --skill subagent-memory -a claude-code
# Install HTML Discussion
npx skills add fourcolors/skills --skill html-discussion -a claude-code
Global (User-Scoped)
Global installations make the skills available across all your local command-line and editor sessions.
npx skills add fourcolors/skills --skill <skill-name> -g
Development & Local Testing
If you are contributing to this pack or testing changes locally:
1. List Discoverable Skills
Scan your local repository tree using the CLI compiler to ensure everything formats correctly:
npx skills add . --list
2. Install Locally for Testing
Verify local compilation and copying by installing directly from your working directory:
npx skills add . --skill <skill-name> -a claude-code