warden-bench

warden-bench is a command for Claude Code from vukkt/token-warden. It costs 28 tokens per session (372 once invoked), scanned A, original, MIT.

A command that runs a fixed benchmark suite for token-warden agents. It compares each run with saved reference results and reports token use, differences, warnings, and environment failures.

In plain words
What is it for?
Use it to benchmark one token-warden agent or all of them, compare current results with frozen and best baselines, and review benchmark costs and warnings.
Why use it?
It makes it possible to check whether an agent's efficiency changes are actually improving results. It also highlights unusually variable tasks and runs affected by service or quota problems.

Command for Claude Code

Written for Claude Code: allowed-tools in frontmatter.

Part of the token-warden plugin — 6 commands, 4 agents, 3 hooks shipped together

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add commands/vukkt/token-warden/warden-bench
Clone the repo
git clone --depth 1 https://github.com/vukkt/token-warden

Made for: Claude Code.

Or install token-warden, the plugin that ships this one along with the rest of its 6 commands, 4 agents, 3 hooks.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for warden-bench

README.md
[![agentmods](https://agentmods.dev/badge/commands/vukkt/token-warden/warden-bench.svg)](https://agentmods.dev/commands/vukkt/token-warden/warden-bench)
Your own site
<a href="https://agentmods.dev/commands/vukkt/token-warden/warden-bench"><img src="https://agentmods.dev/badge/commands/vukkt/token-warden/warden-bench.svg" alt="Measured on agentmods" height="20"></a>
Per session 28 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 372 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00028 $0.00372
Opus 5 $0.00014 $0.00186
Sonnet 5 $0.00006 $0.00074
Haiku 4.5 $0.00003 $0.00037

Measured 6d ago against content hash 300239859ee2, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade A, and why

warden-bench scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 6d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

commands/warden-bench.md · 34 lines

What it actually says

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 task, so it can take several minutes per agent — run it in the background and report progress):

cd "${CLAUDE_SKILL_DIR}/.." && npx tsx src/bench.ts --agent <agent-or-all> <extra flags as given>

When it finishes, report:

  1. Per task: the mean completed tokens, the comparison against the frozen run1 baseline (the printed percentage) and the best figure.
  2. The meta-cost line at the end — and if the overhead warning is present, surface it prominently: it means benchmarking spent more than 10% of the week's collected real-work tokens, and the user should bench less often.
  3. Any per-task runs differ by >25% variance warnings.
  4. If the run prints ENVIRONMENT FAILURE and aborts, report that the benchmark died environmentally (4 consecutive zero-token failed runs — quota exhaustion / API outage) and should be re-run on a fresh quota window; results recorded before the abort are valid.

Never edit run1 values or the database by hand; baselines are frozen by design.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 6d ago First seen · 34 lines · 28 tokens per session scan A 300239859ee2

Subscribe to this mod's changes

warden-bench is a command published in the GitHub repository vukkt/token-warden (13 stars, last pushed 4d ago), licensed MIT. It adds 28 tokens to every session and 372 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.