Skip to content

README.md

Latest commit

 

History

History
178 lines (126 loc) · 6.01 KB

README.md

File metadata and controls

178 lines (126 loc) · 6.01 KB

Dev Experts

Critical, opinionated developer personas as autonomous agents. Claude invokes them automatically based on task context, or you can invoke manually.

Agents

architect - Technical Lead & Architect

Auto-invoked for feature design and implementation planning. Critical, explores alternatives, creates actionable plans.

Workflow:

  • Interrogates requirements
  • Challenges assumptions
  • Explores 2-3 alternatives with trade-off analysis
  • Asks you to choose
  • Creates implementation plan in .claude/experts/plans/[slug].md

Auto-invoked when: Designing features, architecture decisions, implementation planning

devops - Production Detective

Auto-invoked for debugging production issues and infrastructure problems. Methodical and systematic. Assumes the simplest breakage reason first (Occam's Razor).

Phase 0 — Environment Readiness (mandatory before any fix):

  • Git state: destination branch pulled, submodules synced recursively
  • Caches & build artifacts: reset to clean
  • Environment files: validated against examples, dependency versions checked
  • LSP & static analysis: no critical errors in changed files vs baseline
  • Dramatic changes flagged — user asked for explanation if unclear

Workflow (after environment is verified clean):

  • Gathers evidence (symptoms, timing, changes)
  • Forms hypotheses ranked simplest → most complex
  • Tests systematically, escalating complexity only when simpler causes are ruled out
  • Provides diagnostic commands and investigation path

Auto-invoked when: Production issues, mysterious bugs, deployment failures

rust-dev - Rust Purist

Auto-invoked for Rust code review and refactoring. Hunts un-Rusty patterns.

Focus:

  • Ownership & borrowing issues
  • Error handling patterns
  • Type system usage
  • Async correctness
  • Performance optimizations
  • Safety violations

Refactoring Mode: Invoke with "refactor for maintainability" to get comprehensive refactoring plans saved to .claude/plans/[slug]-ref.md

Auto-invoked when: Rust code review, idiom improvements, refactoring requests

python-dev - Pythonista

Auto-invoked for Python code review and refactoring. Hunts un-Pythonic code. Mandates UV and msgspec.

Focus:

  • Type safety (hints, msgspec validation)
  • Async pitfalls
  • Error handling
  • Modern Python features (3.10+)
  • Data structures (dataclasses, TypedDict)
  • Code smells

Mandates: UV for package management, msgspec for validation (not Pydantic)

Refactoring Mode: Invoke with "refactor for maintainability" to get comprehensive refactoring plans saved to .claude/plans/[slug]-ref.md

Auto-invoked when: Python code review, modernization, refactoring requests

cpp-dev - C++ Performance Purist

Auto-invoked for C++ code review and refactoring. HFT mindset. Hunts UB, memory bugs, latency killers.

Focus:

  • Use-after-free, dangling refs, missing RAII
  • Hot path allocations, cache misses, false sharing
  • Memory ordering bugs, ABA problems
  • Integer overflow, bounds violations
  • Concepts over SFINAE, constexpr everywhere

Refactoring Mode: Invoke with "refactor for maintainability" to get comprehensive refactoring plans saved to .claude/plans/[slug]-ref.md

Auto-invoked when: C++ code review, performance optimization, modernization, refactoring requests

reviewer - Grumpy Code Wizard

Auto-invoked for comprehensive code review before commits/PRs. 40 years experience. Reads EVERY line.

Catches:

  • Security holes
  • Race conditions
  • Performance sins (O(n²), N+1 queries)
  • Edge cases (empty inputs, overflow, concurrent access)
  • Maintainability issues

Output: Line-by-line analysis with specific fixes

Auto-invoked when: Pre-merge review, security audit, bug hunting

tester - Testing Specialist

Auto-invoked when writing tests for new features. Writes comprehensive, real-world tests. No fake tests or useless comments.

Workflow:

  • Understands test infrastructure
  • Analyzes changes
  • Identifies test cases (happy path, edge cases, errors)
  • Writes real tests with actual data

Auto-invoked when: Writing tests, improving coverage

Color Scheme

❤️ RED = architect, devops (deciders, coordinators) 💙 BLUE = cpp-dev, python-dev, rust-dev, tester (builders) 💛 YELLOW = reviewer (checker)

Flow

flowchart TD
    USER([USER]) --> architect[/"architect<br/>🔴 Plans & Decides"/]

    subgraph planning [Planning Phase]
        architect -->|"design"| plan["📋 Implementation Plan"]
    end

    subgraph implementation [Implementation Phase]
        plan --> coders
        coders["Language Experts 💙<br/>cpp-dev / python-dev / rust-dev"]
        coders -->|"challenge each other"| coders
    end

    subgraph testing [Testing Phase]
        coders -->|"code ready"| tester["tester 💙<br/>Writes tests"]
    end

    subgraph review [Review Phase]
        tester -->|"tests ready"| reviewer["reviewer 💛<br/>Validates everything"]
    end

    reviewer -->|"issues found"| coders
    reviewer -->|"✅ approved"| done["🚀 Ready"]

    devops["devops 🔴<br/>(when production breaks)"] -.->|"Phase 0: env readiness"| readiness{"Environment<br/>clean?"}
    readiness -->|"yes"| investigate["investigate"]
    readiness -->|"no — fix env first"| readiness
    investigate -.->|"diagnose"| coders
Loading

Phases:

  1. Plan: architect designs feature, explores 2-3 alternatives, creates plan
  2. Implement: Language experts (cpp-dev/python-dev/rust-dev) build it, can challenge each other
  3. Test: tester writes comprehensive tests
  4. Review: reviewer validates code quality, sends back if issues
  5. Ship: Ready for merge

Related Plugins

Bug Hunting: For systematic bug hunting with spec reconstruction and adversarial validation, see bug-hunters. Bug hunters use dev-experts agents (cpp-dev, python-dev) as challengers for false-positive filtering.

Installation

/plugin marketplace add git@github.com:DeevsDeevs/agent-system.git
/plugin install dev-experts@deevs-agent-system