vukkt

Warden Bench — Development skill for Claude Code

Development community

Run the golden benchmark suite for one token-warden agent (or all) and compare results against the frozen run1 and best baselines.

How to install Warden Bench

Installs to ~/.claude/skills/vukkt-token-warden-warden-bench/SKILL.md

Terminal

mkdir -p ~/.claude/skills/vukkt-token-warden-warden-bench && curl -fsSL https://raw.githubusercontent.com/vukkt/token-warden/HEAD/commands/warden-bench.md -o ~/.claude/skills/vukkt-token-warden-warden-bench/SKILL.md

Restart Claude Code, or start a new session, for it to be picked up.

What Warden Bench does

description: Run the golden benchmark suite for one token-warden agent (or all) and compare results against the frozen run1 and best baselines. argument-hint: [--runs N] [--task id] disable-model-invocation: true allowed-tools: Bash(cd:), Bash(npx tsx:)

Run the golden-suite benchmark. The agent (or `all`) and any extra flags are: $ARGUMENTS

If no agent was given, use `all`. Build and run this command (it spawns headless Claude sessions per golden ta

Alternatives in Development

Pool Compare — Compare all Meteora DLMM pools for a token pair by APR, fee/TVL ratio, and volume 722 ★
Cavemem — Frozen — cross-agent persistent memory for coding assistants 674 ★
Caveman Micro — 6-line caveman micro prompt (85 tokens) that outperformed the original 552-token skill 162 ★