convez.ai
MCP ServerDevOps & InfraestruturaIntermediário
pauldatta

Gemini CLI Scanner

pauldatta/gemini-cli-scanner

Audite e otimize seu ecossistema de IA com scoring de maturidade e descoberta de skills

Ver no GitHub
InicialPluginScore · de 0 a 100

MCP Server que audita seu ecossistema de ferramentas de IA (Gemini CLI, Claude Code, Continue, Windsurf, JetBrains AI, OpenCode). Analisa histórico de conversas, cataloga skills e extensões, gera score de maturidade (0-67) com recomendações, e sugere skills reutilizáveis baseado em padrões reais de uso.

Instalação

$npx convez-ai pauldatta/gemini-cli-scanner

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
2
Forks
1
Acompanhando
0
Saúde
57%
Último commit
há 3 meses
Versão
v3.5.8
Issues abertas
15
Contribuidores
2

Casos de uso

  • Diagnosticar maturidade da configuração de ferramentas de IA
  • Descobrir e sugerir skills reutilizáveis a partir de padrões de uso
  • Catalogar MCP servers, extensions, agents e context files
  • Analisar histórico de conversas para extrair conhecimento tribal
  • Gerar relatórios estruturados de governança de IA

Veredito

Ferramenta bem-pensada para governança e auditoria de ecossistemas de IA. Interface TUI intuitiva, modo headless, e geração inteligente de skills a partir de dados reais. Impacto direto em produtividade e estrutura arquitetural de agentes.

README

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

gemini-cli-scanner

Tests npm version Node.js License

Audit your AI coding tool ecosystem — Gemini CLI, Claude Code, Antigravity 2.0 (Desktop, CLI, IDE), Continue, Windsurf, JetBrains AI, and OpenCode. Extract tribal knowledge from how you actually use AI coding tools and surface it as reusable skills, agents, and best practices.

Quick Start

npx gemini-cli-scanner

No install required. Launches an interactive TUI with:

Menu ItemWhat It Does
Quick ScanEnvironment scan, optional repo discovery — no API key needed
Full ScanEverything + AI-powered skill suggestions from your conversation patterns
View ReportScrollable markdown report with t for TOC section-jump navigation
Maturity DashboardScore breakdown (0–67), tier, advisory recommendations
Auth SettingsCredential switching with export hints for session persistence

Headless mode

npx gemini-cli-scanner --skip-suggestions
npx gemini-cli-scanner --repos ~/Code --chat-days 30

What This Does

  1. Catalogs your MCP servers, skills, extensions, agents, policies, and context files
  2. Analyzes conversation history — tools, models, topics, prompt patterns
  3. Scores maturity (0–67) across 11 categories with actionable recommendations
  4. Discovers ecosystems — Gemini CLI, Claude Code, Antigravity (Desktop, CLI, IDE), Continue, Windsurf, JetBrains AI, OpenCode
  5. Suggests skills using a two-stage AI pipeline grounded in your real usage data
  6. Produces a shareable JSON manifest + markdown report (credentials auto-redacted)

Maturity Model

The advisory engine evaluates 11 categories and assigns a maturity tier:

TierScoreWhat It Means
🌱 Getting Started0–33Basic install, minimal config
🔧 Intermediate34–46Active usage with some governance
⚡ Advanced47–60Strong policies, skills, MCP governance
🏆 Expert61–67Full-stack: hooks, extensions, context architecture

📖 Full scoring breakdown and category details →

Install as Agent Skill

Install as a reusable agent skill so your AI assistant knows how to scan on demand:

npx skills add pauldatta/gemini-cli-scanner@scan -g -y

Then just ask: "scan my environment" or "audit my setup".

Install as Extension

For /scan commands inside Gemini CLI:

gemini extensions install https://github.com/pauldatta/gemini-cli-scanner

Partial Environments

The scanner works even when Gemini CLI is not installed. When ~/.gemini is missing:

  • Gemini-specific sub-scans (settings, skills, conversations, memory) are skipped
  • All other AI tool scans (Claude Code, Windsurf, Continue, JetBrains AI, OpenCode) run normally
  • Repository-level scans (GEMINI.md, CLAUDE.md, local .gemini/ configs) proceed as usual
  • The report includes a prominent installation banner and the ecosystem table marks Gemini CLI as "❌ Not installed"

This means your team can collect scan reports from everyone — including engineers who haven't adopted Gemini CLI yet — for a complete picture.

Configure (for AI skill suggestions)

# Option A: Vertex AI
export GOOGLE_CLOUD_PROJECT="your-project"

# Option B: API key
export GOOGLE_API_KEY="your-key"

Without these, the scanner runs fully — it just skips AI-powered skill suggestions.

📖 Skill suggestion pipeline details →

CLI Options

npx gemini-cli-scanner [OPTIONS]

--version, -v         Show version and exit
--gemini-dir PATH     Path to .gemini dir (default: ~/.gemini)
--home-dir PATH       Home directory for Claude scanning (default: ~)
--output-dir PATH     Output directory (default: ./scan-results)
--repos PATH [PATH..] Code repo paths or parent directories to scan
--repo-depth N        Max depth for recursive repo discovery (default: 3)
--chat-days N         Only include conversation data from the last N days
--skip-suggestions    Skip AI skill suggestion (no API key needed)
--json-only           Output JSON only, no markdown report
--include-prompts     Include raw prompts, topics, and project names in output manifest
--skip-update-check   Don't check GitHub for newer versions

Team Dashboard

Aggregate scan reports from your team into a single leadership dashboard:

# Auto-discovers reports in ./scan-results/, ./reports/, or current dir
npx gemini-cli-scanner dashboard

# Or point at a specific folder
npx gemini-cli-scanner dashboard --source ~/team-reports/

Opens a browser dashboard at http://localhost:3847 showing:

SectionWhat It Shows
Overview CardsTeam coverage, average maturity score, top risk
Maturity DistributionExpert/Advanced/Intermediate/Getting Started breakdown
CLI DetectionWhich AI coding tools (Gemini CLI, Claude Code, Windsurf, etc.) are installed per engineer
Toolkit CandidatesSkills and MCP servers shared across engineers — select and assemble into a toolkit spec
Policy ComplianceYOLO gaps, missing deny rules, mode coverage
Tool Chain PatternsBehavioral sequences repeated across the team (automation candidates)
Reporter ScorecardsPer-engineer score, trend arrows, expandable drill-down

How It Works

  1. Engineers run npx gemini-cli-scanner and save their gemini-env-manifest.json to a shared folder
  2. Leader runs npx gemini-cli-scanner dashboard --source ./that-folder/
  3. Dashboard reads all manifests, deduplicates, and builds a cumulative team-state.json
  4. Click Sync to pick up new reports. Click Assemble to generate a toolkit spec from selected candidates

Options

npx gemini-cli-scanner dashboard [OPTIONS]

--source PATH    Folder containing manifest JSON files (auto-discovers if omitted)
--port PORT      Server port (default: 3847)

Output

After running, check scan-results/:

  • gemini-env-manifest.json — Structured data for aggregation
  • gemini-env-report.md — Human-readable report with scores, recommendations, and skills

Privacy & Redaction

Auto-redacted: API keys (AIza..., sk-...), OAuth tokens (ya29...), GitHub PATs (ghp_...), and any field named token, secret, password, or api_key.

Privacy by default (v3.5.6+): The output manifest (gemini-env-manifest.json) contains only aggregate statistics — counts, tool names, model usage, and chain fingerprints. Raw user prompts, thought topics, and project names are not written to disk. The AI skill pipeline uses full data in-memory. Use --include-prompts to include raw data for local analysis. The markdown report always retains full detail as a local-only artifact.

Not touched: Shell history, browser data, or files outside ~/.gemini/, ~/.claude/, and --repos paths.

Documentation

DocDescription
📊 Advisory EngineScoring categories, maturity tiers, detailed point breakdowns
🔍 Scanning ReferenceWhat gets scanned, recursive discovery, ecosystem detection
🤖 Skill SuggestionsTwo-stage AI pipeline, quality standards, methodology
👥 For TeamsCross-team aggregation, maturity benchmarking, enterprise patterns
🌐 Full DocumentationVitePress docs site with search, dark mode, and all guides

Developer Install

git clone https://github.com/pauldatta/gemini-cli-scanner.git
cd gemini-cli-scanner
make          # Interactive TUI
make test     # 263 tests across 12 test files

Contributing

Issues are automatically triaged by Gemini. Use the issue templates for bug reports and feature requests. Collaborators can re-triage by commenting /triage.

License

Apache 2.0

Ver o original no GitHub

Bom para

Relacionados