vbw

VBW Token Efficiency vs Stock Opus 4.6 Agent Teams

Every new capability is shell-only — 85 scripts run as bash subprocesses at zero model token cost. The codebase grew 42% since v1.21.30 while per-request overhead grew just 12% (still 7% below v1.20.0). The Worktree Isolation milestone (41 commits, 6 phases) shipped at near-zero model cost — only +16 reference lines loaded lazily. Dead code removal actually shrank the codebase by 1,665 lines vs the CC Alignment snapshot. 845 bats tests validate the stack.

Analysis reports: v1.30.0 | v1.21.30 | v1.20.0 | v1.10.7 | v1.10.2 | v1.0.99

Category	Stock Agent Teams	VBW	Saving
Base context overhead	10,800 tokens	1,500 tokens	86%
State computation per command	1,300 tokens	200 tokens	85%
Agent coordination (x4 agents)	16,000 tokens	1,200 tokens	93%
Compaction recovery	5,000 tokens	700 tokens	86%
Context duplication (shared files)	16,500 tokens	900 tokens	95%
Agent model cost per phase	$2.78	$1.40	50%
Total coordination overhead	87,100 tokens	12,100 tokens	86%

What this means for your bill: Each phase eliminates ~86% of coordination tokens. For API users, per-phase cost drops from $2.78 to $1.40 (Balanced) or $0.70 (Budget). For subscription plans, ~3x more phases per rate limit cycle — 200% more development capacity for the same price.

Scenario	Without VBW	With VBW (Balanced)	Impact
API: single project (10 phases)	~$28	~$14	~$14 saved
API: active dev (20 phases/mo)	~$56/mo	~$28/mo	$28/mo saved ($336/yr)
API: heavy dev (50 phases/mo)	~$139/mo	~$70/mo	$69/mo saved ($828/yr)
API: team (100 phases/mo)	~$278/mo	~$139/mo	$139/mo saved ($1,668/yr)
Pro / Max subscription	baseline capacity	~3x phases per cycle	200% more work done

Budget profile ($0.70/phase) doubles API savings. Quality profile ($2.80/phase) matches stock cost but adds V2/V3 enforcement at zero premium. Based on current API pricing.

Manifesto

VBW is open source because the best tools are built by the people who use them.

This project exists to make AI coding better for everyone, and "everyone" means exactly that.

For absolute beginners: VBW may look intimidating, especially if you've never used Claude Code, but it is, in fact, incredibly easy to use. And your results will be significantly better than using an IDE with a chatbot.

For seasoned developers: Everything is configurable — effort profiles, autonomy levels, model routing, verification depth, work presets — all exposed as a control surface, not hidden behind prompts. The beginners get guardrails; you get the switches behind the guardrails.

For contributors: VBW is a living project. The plugin system, the agents, the verification pipeline - all of it is open to improvement. If you've found a better way to plan, build, or verify code with Claude, bring it. File an issue, open a PR, or just show up and share what you've learned. Every contribution makes the next person's experience better.

Join the Discord -- whether you want to help build VBW or just want VBW to help you build.

What Is This

Platform: macOS and Linux only. Windows is not supported natively — all hooks, scripts, and context blocks require bash. If you're on Windows, run Claude Code inside WSL.

Inspired by Ralph and Get Shit Done, however, an entirely new architecture.

VBW is a Claude Code plugin that bolts an actual development lifecycle onto your vibe coding sessions.