Deep Dive – Análise Multi-Agent Rigorosa
nelsonwerd/deep-dive.skill
Investigação multi-agent paralela com síntese, red-team e veredito confiável
Ver no GitHubSkill que orquestra 4–6 agentes especializados em paralelo para análise profunda de codebases, estratégias, designs e pesquisas abertas. Executa síntese, verificação adversarial e produz relatório estruturado com confiança quantificada (1–10). Ideal para decisões de alto risco.
Instalação
npx convez-ai nelsonwerd/deep-dive.skill1 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
- Auditoria técnica de codebases complexas antes de ship
- Avaliação rigorosa de estratégias e arquiteturas de sistema
- Pesquisa aberta com verificação de claims contraditórios
- Due diligence para decisões financeiras ou críticas
Veredito

Skill robusto para análise investigativa em larga escala, mas explicitamente token-intensivo — requer planejamento em planos menores. Altamente recomendado para equipas que precisam de vereditos confiáveis em decisões de alto risco.
README
A tradução para o português deste README ainda está na fila.
deep-dive
Rigorous multi-agent deep-dive analysis for Claude Code. Point it at a codebase, a strategy or decision system, a design, or an open research question and it runs a structured investigation instead of giving a one-shot answer.
What it does
For a complex, investigative task, the skill runs a multi-phase pipeline:
- Parallel specialists — deploys 4–6 subagents at once, each in its own lane.
- Synthesis — one agent reads every specialist output, cross-checks claims, and resolves contradictions.
- Follow-up verification — single-claim agents re-check load-bearing or single-sourced numbers.
- Red-team — an adversarial reviewer tries to break the conclusions.
- Executive briefing — a plain-English verdict with an honest 1–10 confidence rating.
It scales down for narrow scope (2–3 lanes) and up for broad, multi-domain work (6 lanes). By default it runs in pure research mode — it writes markdown findings and changes no code unless you explicitly ask.
Four built-in variants live in deep-dive/references/: codebase audit, strategy / system evaluation, design evaluation, and open-ended research.
Heads up — this skill is token-hungry by design. A full run fans out 4–6 specialist agents (each writing thousands of words), then synthesis, follow-up verification, a red-team pass, and a briefing — easily 10+ agent calls and tens of thousands of tokens for a single analysis. That's the right trade for a high-stakes call (a ship/no-ship decision, real money on the line), and a great fit on a Claude Max plan or any setup where you're not token-constrained. On a smaller plan, reach for it deliberately: use the Scale heuristics in
SKILL.md(2–3 lanes for narrow scope, skip the red-team for low-stakes work), or just ask for a single-pass review instead.
Install
Skills live in ~/.claude/skills/ (all projects) or .claude/skills/ (a single project). Put the deep-dive/ folder in either one.
From the packaged file:
mkdir -p ~/.claude/skills
unzip deep-dive.skill -d ~/.claude/skills/
Or from a clone of this repo:
git clone https://github.com/nelsonwerd/deep-dive.skill.git
mkdir -p ~/.claude/skills
cp -r deep-dive.skill/deep-dive ~/.claude/skills/
No restart needed — Claude Code detects it in-session. Verify with /skills, or just ask Claude what skills are available, and confirm deep-dive is listed.
Works in Claude and Codex
This follows the open Agent Skills standard, so the same SKILL.md works in Claude and OpenAI Codex:
| You are… | Tool | How |
|---|---|---|
| Non-technical | Claude app (claude.ai / desktop) | Upload deep-dive.skill (a zip) in the app's Skills / Capabilities settings → Agent Skills docs |
| Developer | Claude Code | the install above (drop deep-dive/ in ~/.claude/skills/) |
| Developer | OpenAI Codex | copy deep-dive/SKILL.md (+ references/) into .agents/skills/deep-dive/ in your repo, or ~/.agents/skills/deep-dive/ globally → Codex skills docs |
| Anyone | any agent | it's just instructions — open SKILL.md and point your agent at it |
Menu names/commands drift between versions — the linked docs are the source of truth. One caveat specific to deep-dive: its parallel multi-agent orchestration is native to Claude Code; in Codex it runs as a single-agent, guided version of the same playbook (the method carries; the parallelism doesn't).
Use it
- Manually: type
/deep-diveand describe the target. - Automatically: Claude invokes it on its own when a task looks like a thorough audit, rigorous analysis, or comprehensive review.
Examples:
- "Do a deep dive on this codebase before I ship it."
- "Evaluate whether this strategy actually holds up under scrutiny."
- "Is this the right architecture? Review it thoroughly."
What's in this repo
deep-dive/— the skill itself (SKILL.md+ reference playbooks). This is what you install.deep-dive.skill— the same folder, zipped, for a one-step download.