warden-select

warden-select is a command for Claude Code from vukkt/token-warden. It costs 29 tokens per session (631 once invoked), scanned A, original, MIT.

A benchmarking command for deciding whether candidate rules for an agent's token-warden memory should be enabled or removed. It tests those rules against a shared golden suite, a fixed set of evaluation cases.

In plain words
What is it for?
Use it to evaluate up to three candidate rules per run, perform re-audits, and recompile the agent's memory.
Why use it?
It provides measurements before changing the agent's memory, while allowing extra checks when results are close or noisy.

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-select
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-select

README.md
[![agentmods](https://agentmods.dev/badge/commands/vukkt/token-warden/warden-select.svg)](https://agentmods.dev/commands/vukkt/token-warden/warden-select)
Your own site
<a href="https://agentmods.dev/commands/vukkt/token-warden/warden-select"><img src="https://agentmods.dev/badge/commands/vukkt/token-warden/warden-select.svg" alt="Measured on agentmods" height="20"></a>
Per session 29 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 631 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 $0.00029 $0.00631
Opus 5 $0.00015 $0.00316
Sonnet 5 $0.00006 $0.00126
Haiku 4.5 $0.00003 $0.00063

Measured 5d ago against content hash 558a671eaee7, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-05, from the pricing page.

Security

Grade A, and why

warden-select 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 5d 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-select.md · 52 lines

What it actually says

Run the token-warden selector. The agent and any extra flags are: $ARGUMENTS

This spends real benchmark tokens (a shared baseline suite, one suite per candidate — max 3 per invocation — one re-audit, plus possible variance top-ups), and takes several minutes per configuration. Run it in the background and report progress:

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

--top-up N (default 1) is how many ORDINARY measurement rounds a decision whose verdict lands within noise of the bar may buy. Each round is one more suite pass of the measured configuration, placed by variance. 0 disables the top-up entirely.

--uniform-top-up replaces the Neyman variance-proportional top-up with one full uniform suite pass (same budget) — the control arm when benchmarking the allocation strategy itself.

--retention-rounds N (0-2, default 2) caps the EXTRA measurement rounds a RE-AUDIT may buy before de-activating a rule with a banked margin. A re-audit whose noise band is wide relative to what the rule has already been shown to earn buys more evidence rather than deciding on the noisy draw; a decisive measurement, a regression, and every candidate promotion are unaffected. 0 restores the pre-v0.43.0 single-top-up behaviour and is the control arm.

When it finishes, report:

  1. Each decision: rule id, ACTIVE/EVICTED, measured delta vs. context rent, the advisory dollar translation when shown, and any REGRESSION / topped-up / LOW-CONFIDENCE annotations.
  2. The compiled memory path and new ruleset version.
  3. If a previously active rule was evicted on re-audit, say so explicitly — that is mandatory eviction working, not a malfunction.
  4. If the run prints ABORTED: environment failure (and exits non-zero), report that NO verdict was recorded — the measurement died environmentally (quota exhaustion / API outage), the rule is still queued as a candidate, and the fix is to re-run /warden-select on a fresh quota window. Never describe an abort as an eviction.

The "≈$/run" and "$/week" figures are advisory only — the keep/evict verdict is decided on raw tokens, never dollars. Never edit rules or MEMORY.md by hand; the ledger is the source of truth.

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. 5d ago First seen · 52 lines · 29 tokens per session scan A 558a671eaee7

Subscribe to this mod's changes

warden-select is a command published in the GitHub repository vukkt/token-warden (13 stars, last pushed 3d ago), licensed MIT. It adds 29 tokens to every session and 631 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.