phpstan-runner

使用智能代理委托运行PHPStan静态分析并修复错误。当用户要求时使用:- 运行静态分析 - 修复PHPStan错误 - 检查代码质量 - 分析类型错误委托给运行代理(haiku)执行,修复代理(sonnet)修复。自动在运行和修复之间循环,直到分析通过或需要人工干预。

📖 中文导读:使用智能代理委托运行PHPStan静态分析并修复错误。当用户要求时使用:- 运行静态分析 - 修复PHPStan错误 - 检查代码质量 - 分析类型错误委托给运行代理(haiku)执行,修复代理(sonnet)修复。自动在运行和修复之间循环,直到分析通过或需要人工干预。 以下正文为该技能的原始 SKILL.md 内容(英文),上方安装命令可直接使用。

name: phpstan-runner description: | Run PHPStan static analysis and fix errors using intelligent agent delegation. Use when user requests to: - Run static analysis - Fix PHPStan errors - Check code quality - Analyze type errors Delegates to runner agent (haiku) for execution and fixer agent (sonnet) for fixes. Automatically cycles between run and fix until analysis passes or human intervention needed. allowed-tools: Task

PHPStan Runner Skill

This skill provides intelligent PHPStan static analysis and fixing through specialized agent delegation.

Agent Delegation Strategy

This skill delegates to specialized agents via the Task tool:

  1. php-qa-ci_phpstan-runner agent (haiku model) - Runs analysis and parses results
  2. php-qa-ci_phpstan-fixer agent (sonnet model) - Analyzes and fixes errors
  3. Escalation - Uses opus model or asks human for stubborn issues

Workflow

When User Says: "Run PHPStan"

  1. Launch runner agent:

    Use Task tool:
      description: "Run PHPStan analysis"
      subagent_type: "php-qa-ci_phpstan-runner"
      prompt: "Run PHPStan static analysis and provide summary"
    
  2. Receive runner output with log location

  3. If errors detected:

    • Launch fixer agent:
      Use Task tool:
        description: "Fix PHPStan errors"
        subagent_type: "php-qa-ci_phpstan-fixer"
        prompt: "Fix errors in log: {log_path}"
      
  4. After fixes applied, re-run via runner agent

  5. Repeat cycle until:

    • Analysis passes → Success
    • Same errors persist 2+ times → Escalate to opus or human
    • User intervention needed → Ask user

When User Says: "Fix the PHPStan errors"

  1. Check if recent log exists in var/qa/phpstan_logs/

  2. If log found:

    • Launch fixer agent directly with log path
  3. If no log:

    • Launch runner agent first to generate log
    • Then launch fixer agent

Escalation Triggers

Launch opus model or ask human when:

  • Fixer agent reports "cannot fix" for same error 2+ times
  • Architecture questions arise (design patterns, type hierarchies)
  • User explicitly requests explanation of errors

Runner Agent Reference

The phpstan-runner agent (haiku model) handles:

  • PHPStan execution with proper configuration
  • Log parsing for error patterns
  • Concise summary generation

See .claude/agents/php-qa-ci_phpstan-runner.md for agent implementation details.

Fixer Agent Reference

The phpstan-fixer agent (sonnet model) handles:

  • Log file discovery and parsing
  • Error grouping by pattern
  • Fix implementation for common PHPStan issues
  • Verification that fixes resolve issues

See .claude/agents/php-qa-ci_phpstan-fixer.md for agent implementation details.