The Longform Guide

Token Optimization

Effective token management is critical for long coding sessions. ECC provides several strategies:

Model Selection

Match task complexity with the right model tier:

Model	Best For	Token Cost
Opus	Architecture, code review, security analysis	Highest
Sonnet	Implementation, build fixing, TDD	Medium
Haiku	Documentation lookup, simple queries	Lowest

System Prompt Slimming

Keep your CLAUDE.md focused. Each line costs tokens on every message. Use skills (loaded on-demand) instead of cramming everything into the system prompt.

Background Processes

Use the /aside command for side conversations that don’t pollute your main context. Use /checkpoint to save state before long operations.

Memory Persistence

Hooks can automatically save and restore context across sessions:

Session save (/save-session) - Persists current state
Session resume (/resume-session) - Restores previous state
Auto-compact - PreCompact hook extracts key context before compaction

Continuous Learning

ECC includes patterns for self-improving workflows:

Pattern extraction - After sessions, extract reusable patterns into skills
Skill evolution - Skills improve over time based on usage data
Instinct system - Export/import learned behaviors across projects

Commands: /learn, /learn-eval, /instinct-export, /instinct-import

Verification Loops

Quality assurance through automated evaluation:

Checkpoint evals - Validate at specific milestones
Continuous evals - Run checks throughout development
Grader types - Automated scoring of code quality
pass@k metrics - Statistical confidence in outputs

Commands: /eval, /quality-gate

Parallelization

Scale development across multiple instances:

Git worktrees - Isolated workspaces for parallel tasks
Cascade method - Chain agent outputs through sequential steps
Multi-agent orchestration - Coordinate multiple Claude instances

Commands: /devfleet, /multi-plan, /multi-backend, /multi-frontend