Performance & Cost

Token Efficiency

Adaptive Planning

Tasks are planned based on complexity:

Task Type	Planning	Tokens
Fix typo	Skip	~200
Add field	Quick plan	~500
New module	Full plan	~5,000

Simple tasks skip unnecessary planning.

Unified Review

One agent handles three review types:

1. Spec compliance — Does it meet requirements?
2. Code quality — Is it clean and tested?
3. Architecture fit — Does it follow patterns?

Single pass = fewer tokens.

Cost Scaling

Tasks	Tokens	Est. Cost
1	~5k	~$0.08
2	~10k	~$0.15
3	~15k	~$0.23
4	~20k	~$0.30

Linear scaling. More tasks don't multiply costs.

Full vs Lite Mode

Mode	Model	Tokens	Cost
Full	Sonnet/Opus	~15-25k	Higher
Lite	Haiku	~5-10k	60-80% less

Lite mode is ideal for:

• Module scaffolding
• Quick components
• Rapid iterations

Full mode is ideal for:

• Complex features
• PR reviews
• Production code

Feature Comparison

Feature	Description
Adaptive planning	Adjusts to task complexity
Unified review	3 checks in 1 pass
TDD verification	Proof-based, not trust-based
Checkpoints	Git save/restore
Session memory	Remembers decisions
Cost tracking	Token metrics per session
Multi-platform	Claude Code, Cursor, VS Code, Windsurf, Codex
Framework packs	Next.js, React, Vue, SvelteKit, Angular, Astro, Nuxt

Example: User Authentication

Tasks:

1. Create auth service
2. Create auth hook
3. Create login component
4. Create tests
5. Review code

Token breakdown:

Phase	Full	Lite
Planning	~1,500	~500
Building	~8,000	~3,000
Review	~3,000	~1,000
Total	~12,500	~4,500

Cost estimate:

• Full (Sonnet): ~$0.19
• Lite (Haiku): ~$0.02

Verification

TDD Verification

Tests are proof-based:

FAIL src/discount.test.ts
  Expected: 90
  Received: undefined

[Evidence of RED phase captured]

Not "I ran the test" — actual output.

Review Verification

Clear verdicts:

Spec Compliance: PASS
Code Quality: PASS
Architecture: FAIL

Verdict: Requires Changes

Session Metrics

After each session:

Session Summary
───────────────
Tasks: 5/5 complete
Tokens: 45,230 (~$0.68)
Time: 12m 34s
Commits: 7
Files: 23

Agent Breakdown:
  @planner: 8,200 tokens
  @builder: 28,400 tokens
  @reviewer: 8,630 tokens

Track costs as you work.