Everything Claude Code Methodology

Performance optimization system for AI agent harnesses, adapted from the Everything Claude Code project. Integrates specialized subagents, skills, continuous learning, security scanning, and research-first development workflows into babysitter process definitions.

Core Principles

1. **Research-First Development**: Always investigate existing solutions before writing code 2. **TDD Enforcement**: Red-Green-Refactor is mandatory; 80% coverage threshold 3. **Multi-Dimensional Review**: 4-axis code review with confidence-gated reporting (>= 80%) 4. **AgentShield Security**: 5-category security scanning with 102 static analysis rules 5. **Continuous Learning**: Auto-extract patterns from sessions and convert to reusable skills 6. **Context Engineering**: Dynamic context injection, mode switching, strategic compaction 7. **Evidence-Based Verification**: All claims backed by exit codes, coverage numbers, or test results

Process Files (5)

Agents (8)

Skills (8)

Key Patterns

• **Remediation Loop**: Review findings trigger refactor-cleaner, re-review, max 2 cycles
• **Coverage Convergence**: Iterative test addition until 80% threshold met (max 3 iterations)
• **Parallel Execution**: Code review and security scan run concurrently
• **Breakpoints**: Plan review, coverage threshold, critical findings, deployment approval
• **Cascade Execution**: Topological build order for dependent services
• **Worktree Parallelism**: Git worktree isolation for independent service builds

Attribution

Adapted from Everything Claude Code by Affaan M.