prodscope

Prodscope

Operational-first agentic plugin: a Builder proposes changes with explicit verification plans, and a Breaker (senior SRE mindset) reviews for production risks before anything is treated as done.

Why Prodscope?

AI coding assistants ship code fast but often ignore production realities — missing indexes, race conditions, silent error swallowing, no observability. Prodscope adds a structured Builder-Breaker loop that runs automatically, so every change is challenged before it lands.

Install

One-liner (all platforms)

curl -fsSL https://raw.githubusercontent.com/ashokdudhade/prodscope/main/install.sh | bash

This copies skills, commands, rules, and hooks into your current project. No global install, no dependencies.

Platform-specific

Install only the files you need:

# Claude Code only
curl -fsSL https://raw.githubusercontent.com/ashokdudhade/prodscope/main/install.sh | bash -s -- --claude-only

# Cursor only
curl -fsSL https://raw.githubusercontent.com/ashokdudhade/prodscope/main/install.sh | bash -s -- --cursor-only

# OpenAI Codex only
curl -fsSL https://raw.githubusercontent.com/ashokdudhade/prodscope/main/install.sh | bash -s -- --codex-only

Claude Code (native plugin install)

/plugin marketplace add ashokdudhade/prodscope
/plugin install prodscope@prodscope-marketplace

This is the recommended method for Claude Code — the plugin auto-updates and integrates with the plugin system.

Other native installs

Platform	Method
Cursor	Install from Cursor Marketplace, or Settings → Rules → Remote Rule → paste ashokdudhade/prodscope
Codex	Copy .codex-plugin/, .agents/, AGENTS.md, and skills/ into your project root

Other options

# Preview what would be installed (no changes made)
curl -fsSL https://raw.githubusercontent.com/ashokdudhade/prodscope/main/install.sh | bash -s -- --dry-run

# Clone and install from local copy
git clone https://github.com/ashokdudhade/prodscope.git /tmp/prodscope
cd your-project && /tmp/prodscope/install.sh

Usage

Type /prodscope:build <your task> in Claude Code or Cursor. Codex runs the loop automatically.

/prodscope:build "add rate limiting to the API"
  │
  ├─ Phase 1: Builder
  │   ├─ Implements smallest coherent diff
  │   └─ Attaches HEALTH_CHECK (smoke steps, signals, rollback)
  │
  ├─ Phase 2: Breaker
  │   ├─ SRE review: async races, DB indexes, memory, security
  │   ├─ REJECT → loops back to Builder with required fixes (max 3 cycles)
  │   └─ ACCEPT → proceeds to verification
  │
  ├─ Phase 3: Verification
  │   ├─ Runs HEALTH_CHECK smoke steps with evidence
  │   └─ Failure → loops back to Builder
  │
  └─ Phase 4: Reliability score
      ├─ Rates change 1–10 (tests, observability, errors, ops, security)
      └─ Logs outcome to .prodscope/logs/

The entire loop runs in a single response — no manual handoff between phases.

Commands

Command	Platform	What it does
/prodscope:build <task>	Claude Code, Cursor	Full Builder → Breaker → Verify → Score loop
/prodscope:verify	Claude Code, Cursor	Post-build verification with MCP telemetry (Sentry, Datadog)

MCP integration

mcp-config.json is a placeholder for Sentry, Datadog, or other MCP servers. Add entries per vendor docs and inject API keys via environment variables — never commit secrets. See commands/verify.md for the /prodscope:verify operator template.

What gets installed