Skip to content

Configuration Recipes

Jump to bottom
Z-M-Huang edited this page Mar 10, 2026 · 1 revision

Configuration Recipes

Copy-paste configurations for common setups. Each recipe includes both VCP and Dev Buddy config files.

Recipe 1: Solo Developer — React + Express

Scenario: Single developer, full-stack web app, Claude Code subscription only (no API keys).

VCP Config (.vcp/config.json)

{
  "version": "1.0",
  "scopes": {
    "web-frontend": true,
    "web-backend": true,
    "database": false
  },
  "compliance": [],
  "frameworks": ["react", "express"],
  "exclude": ["node_modules/**", "dist/**", "build/**", "coverage/**"],
  "severity": "medium",
  "ignore": [],
  "pluginRoot": "/home/user/.claude/plugins/cache/vcp/vcp/0.3.0"
}

Why these choices:

  • database: false — Unless you write raw SQL. If you use Prisma/Sequelize with typed queries, the backend rules already cover data access.
  • compliance: [] — No regulatory requirements for a personal project.
  • severity: "medium" — Catches the important issues without excessive noise.

Dev Buddy Pipeline (~/.vcp/dev-buddy.json)

{
  "feature_pipeline": [
    { "type": "requirements", "provider": "anthropic-subscription", "model": "opus" },
    { "type": "planning", "provider": "anthropic-subscription", "model": "opus" },
    { "type": "plan-review", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "implementation", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "code-review", "provider": "anthropic-subscription", "model": "opus" }
  ],
  "bugfix_pipeline": [
    { "type": "rca", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "rca", "provider": "anthropic-subscription", "model": "opus" },
    { "type": "implementation", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "code-review", "provider": "anthropic-subscription", "model": "opus" }
  ],
  "max_iterations": 5
}

Why this pipeline:

  • Fewer stages (5 feature, 4 bugfix). Faster iteration for a solo dev.
  • One plan review, one code review. Still catches issues without 3-review redundancy.
  • Opus for requirements/planning, sonnet for implementation.
  • max_iterations: 5 -- escalate quickly since you're the only decision-maker.

Recipe 2: Team with Compliance (GDPR + Backend Focus)

Scenario: Team SaaS platform, EU user data, GDPR compliance required, shared config.

VCP Config (.vcp/config.json)

{
  "version": "1.0",
  "scopes": {
    "web-frontend": true,
    "web-backend": true,
    "database": true
  },
  "compliance": ["gdpr"],
  "frameworks": ["react", "express", "postgresql", "prisma"],
  "exclude": ["node_modules/**", "dist/**", "build/**", "coverage/**", "migrations/**"],
  "severity": "medium",
  "ignore": [],
  "pluginRoot": "/home/user/.claude/plugins/cache/vcp/vcp/0.3.0"
}

Why these choices:

  • database: true — Direct database interactions with PostgreSQL need schema security checks.
  • compliance: ["gdpr"] — Legal requirement for EU user data. Adds PII handling, data deletion, and consent rules.
  • migrations/** excluded — Migration files are generated; scanning them produces false positives.

Dev Buddy Pipeline (~/.vcp/dev-buddy.json)

{
  "feature_pipeline": [
    { "type": "requirements", "provider": "anthropic-subscription", "model": "opus", "review_gate": true },
    { "type": "planning", "provider": "anthropic-subscription", "model": "opus", "review_gate": true },
    { "type": "plan-review", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "plan-review", "provider": "anthropic-subscription", "model": "opus" },
    { "type": "implementation", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "code-review", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "code-review", "provider": "anthropic-subscription", "model": "opus" }
  ],
  "bugfix_pipeline": [
    { "type": "rca", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "rca", "provider": "anthropic-subscription", "model": "opus" },
    { "type": "plan-review", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "implementation", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "code-review", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "code-review", "provider": "anthropic-subscription", "model": "opus" }
  ],
  "max_iterations": 10
}

Why this pipeline:

  • review_gate: true on requirements and planning. Team reviews the user story and plan before automated reviews start.
  • Two plan reviews + two code reviews. More passes for compliance-sensitive code.
  • Bug-fix includes plan-review to validate the RCA diagnosis before fixing.

Recipe 3: Budget-Conscious — Minimize API Usage

Scenario: Multi-AI diversity on a budget. Subscription handles most work, API only for the final code review.

Dev Buddy Presets (~/.vcp/ai-presets.json)

{
  "version": "2.0",
  "presets": {
    "anthropic-subscription": {
      "type": "subscription",
      "name": "Anthropic Subscription"
    },
    "openrouter-budget": {
      "type": "api",
      "name": "OpenRouter Budget",
      "protocol": "anthropic",
      "base_url": "https://openrouter.ai/api",
      "api_key": "sk-or-...",
      "models": ["claude-sonnet-4-5-20250514"],
      "timeout_ms": 300000
    }
  }
}

Dev Buddy Pipeline (~/.vcp/dev-buddy.json)

{
  "feature_pipeline": [
    { "type": "requirements", "provider": "anthropic-subscription", "model": "opus" },
    { "type": "planning", "provider": "anthropic-subscription", "model": "opus" },
    { "type": "plan-review", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "plan-review", "provider": "anthropic-subscription", "model": "opus" },
    { "type": "implementation", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "code-review", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "code-review", "provider": "anthropic-subscription", "model": "opus" },
    { "type": "code-review", "provider": "openrouter-budget", "model": "claude-sonnet-4-5-20250514" }
  ],
  "bugfix_pipeline": [
    { "type": "rca", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "rca", "provider": "anthropic-subscription", "model": "opus" },
    { "type": "implementation", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "code-review", "provider": "anthropic-subscription", "model": "opus" },
    { "type": "code-review", "provider": "openrouter-budget", "model": "claude-sonnet-4-5-20250514" }
  ],
  "max_iterations": 10
}

Why this works:

  • Subscription handles 90% of the work (no per-token cost).
  • API preset used only for the final code review, so you get one API call per pipeline run.
  • The API review runs as a separate subprocess. It's actually a different instance reviewing the code, not Claude reviewing its own output in the same context.

Recipe 4: Maximum Rigor — Multi-Provider with Codex Final Gate

Scenario: Enterprise team, Codex CLI as independent final reviewer, willing to spend on API costs.

Dev Buddy Presets (~/.vcp/ai-presets.json)

{
  "version": "2.0",
  "presets": {
    "anthropic-subscription": {
      "type": "subscription",
      "name": "Anthropic Subscription"
    },
    "anthropic-api": {
      "type": "api",
      "name": "Anthropic API Direct",
      "base_url": "https://api.anthropic.com",
      "api_key": "sk-ant-...",
      "models": ["claude-sonnet-4-5-20250514", "claude-opus-4-20250514"]
    },
    "codex-cli": {
      "type": "cli",
      "name": "OpenAI Codex CLI",
      "command": "codex",
      "args_template": "exec --full-auto --model {model} -o {output_file} --output-schema {schema_path} {prompt}",
      "one_shot_args_template": "exec --full-auto --model {model} \"{prompt}\"",
      "models": ["o3", "o4-mini"],
      "timeout_ms": 1200000
    }
  }
}

Dev Buddy Pipeline (~/.vcp/dev-buddy.json)

{
  "feature_pipeline": [
    { "type": "requirements", "provider": "anthropic-subscription", "model": "opus" },
    { "type": "planning", "provider": "anthropic-subscription", "model": "opus", "review_gate": true },
    { "type": "plan-review", "provider": "anthropic-subscription", "model": "sonnet", "parallel": true },
    { "type": "plan-review", "provider": "anthropic-subscription", "model": "opus", "parallel": true },
    { "type": "plan-review", "provider": "codex-cli", "model": "o3" },
    { "type": "implementation", "provider": "anthropic-subscription", "model": "sonnet", "phased_reviews": [
      { "provider": "anthropic-subscription", "model": "sonnet" }
    ]},
    { "type": "code-review", "provider": "anthropic-subscription", "model": "sonnet", "parallel": true },
    { "type": "code-review", "provider": "anthropic-api", "model": "claude-opus-4-20250514", "parallel": true },
    { "type": "code-review", "provider": "codex-cli", "model": "o3" }
  ],
  "bugfix_pipeline": [
    { "type": "rca", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "rca", "provider": "anthropic-subscription", "model": "opus" },
    { "type": "plan-review", "provider": "codex-cli", "model": "o3" },
    { "type": "implementation", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "code-review", "provider": "anthropic-subscription", "model": "sonnet", "parallel": true },
    { "type": "code-review", "provider": "anthropic-api", "model": "claude-opus-4-20250514", "parallel": true },
    { "type": "code-review", "provider": "codex-cli", "model": "o3" }
  ],
  "max_iterations": 10,
  "max_phased_iterations": 3,
  "review_interval": 1
}

Why this pipeline:

  • Parallel plan reviews (Sonnet + Opus) run at the same time. Codex runs after as a sequential gate. Different AI families catch different things.
  • Phased reviews on implementation, so each step gets checked before the next starts.
  • Three code reviews: subscription Sonnet (fast), API Opus (deep), Codex (different AI family entirely).
  • review_gate: true on planning so a human signs off before the expensive stages.

Recipe 5: Bug-Fix-Only Workflow

Scenario: Existing codebase, mostly bug fixing, want fast RCA turnaround.

Dev Buddy Pipeline (~/.vcp/dev-buddy.json)

{
  "feature_pipeline": [
    { "type": "requirements", "provider": "anthropic-subscription", "model": "opus" },
    { "type": "planning", "provider": "anthropic-subscription", "model": "opus" },
    { "type": "plan-review", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "implementation", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "code-review", "provider": "anthropic-subscription", "model": "opus" }
  ],
  "bugfix_pipeline": [
    { "type": "rca", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "rca", "provider": "anthropic-subscription", "model": "opus" },
    { "type": "rca", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "plan-review", "provider": "anthropic-subscription", "model": "opus" },
    { "type": "implementation", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "code-review", "provider": "anthropic-subscription", "model": "sonnet" },
    { "type": "code-review", "provider": "anthropic-subscription", "model": "opus" }
  ],
  "max_iterations": 5,
  "rca_review_gate": true
}

Why this pipeline:

  • Three RCA stages give you more diagnostic angles on the root cause.
  • rca_review_gate: true -- after all RCAs run and consolidate, you approve the diagnosis before the fix starts.
  • Plan review with opus to validate the consolidated RCA + fix plan.
  • max_iterations: 5 -- bug fixes should converge fast; escalate if they don't.
  • Feature pipeline is minimal (kept as a fallback).

Recipe 6: Chatroom for Architecture Decisions

Scenario: Get different AI models to argue about architecture before you commit to a direction.

Dev Buddy Chatroom (~/.vcp/dev-buddy-chatroom.json)

{
  "participants": [
    { "preset": "openrouter", "model": "claude-sonnet-4-5-20250514" },
    { "preset": "openai-api", "model": "gpt-4.1" },
    { "preset": "minimax", "model": "MiniMax-M2.5" }
  ],
  "max_rounds": 3
}

Why these choices:

  • Three external participants + Claude (built-in) = 4 AI opinions.
  • Different AI families (Claude via API, GPT, MiniMax) so they don't all agree for the same reasons.
  • max_rounds: 3 -- enough for convergence without dragging on.

Usage:

/dev-buddy-chatroom Should we use a microservice or monolith for the payment system?

Each participant gives independent analysis in Round 1, then they respond to each other in subsequent rounds. Claude synthesizes the results at the end.

Mixing and Matching

These are starting points. Common tweaks:

Want Modify
Faster pipelines Remove review stages, use sonnet everywhere
Cheaper pipelines Use subscription for everything, no API presets
More confidence Add more code-review stages, use opus for reviews
User checkpoints Add "review_gate": true on requirements/planning
Step-by-step verification Add "phased_reviews" on implementation
Batch verification Set "review_interval": 3 or higher
Quick escalation Lower max_iterations to 3-5
More autonomy Raise max_iterations to 15-20

See Also

VCP Repository | Report a Problem | Propose a Standard | Apache License 2.0 | English | 中文

VCP Wiki

Guides

VCP Plugin

Dev Buddy Plugin

MCP Doc Plugin

Standards

Project

VCP Wiki (中文)

指南

VCP 插件

Dev Buddy 插件

MCP Doc 插件

标准

项目

Clone this wiki locally