run-evals

run-evals is a command for Claude Code from tyejcoleman/tokenroom. It costs 17 tokens per session (326 once invoked), scanned A, original, Apache-2.0.

A command for running behavioral evaluation suites on changes that may affect an agent’s behavior. The suites compare agent runs and grade evidence from their generated artifacts.

In plain words
What is it for?
Use it after changing skill or policy wording, decision semantics, or handoff rendering, choosing the appropriate evaluation harness and recording the results.
Why use it?
It provides evidence for whether wording or policy changes actually alter behavior, instead of relying only on the agent’s own report.

Command for Claude Code

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/tyejcoleman/tokenroom/run-evals
Clone the repo
git clone --depth 1 https://github.com/tyejcoleman/tokenroom

Made for: Claude Code.

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 run-evals

README.md
[![agentmods](https://agentmods.dev/badge/commands/tyejcoleman/tokenroom/run-evals.svg)](https://agentmods.dev/commands/tyejcoleman/tokenroom/run-evals)
Your own site
<a href="https://agentmods.dev/commands/tyejcoleman/tokenroom/run-evals"><img src="https://agentmods.dev/badge/commands/tyejcoleman/tokenroom/run-evals.svg" alt="Measured on agentmods" height="20"></a>
Per session 17 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 326 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 $0.00017 $0.00326
Opus 5 $0.00009 $0.00163
Sonnet 5 $0.00003 $0.00065
Haiku 4.5 $0.00002 $0.00033

Measured 4d ago against content hash 02b48661aeb4, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

run-evals 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 4d 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.

.claude/commands/run-evals.md · 24 lines

What it actually says

Run the behavioral evals for: $ARGUMENTS

Per ADR-9, any change claiming to affect model behavior (stamp wording, SKILL.md policy, fit_check verdict semantics, handoff rendering) needs eval evidence before merging.

  1. Pick the harness: eval/ (v0 — cheap single-shot planning probes; fastest iteration), eval/v1/ (execution-level with live simulated budgets; the headline numbers), eval/v2-continuity/ (post-compaction resume). Read the harness README first.
  2. Use the harness's setup scripts to generate cells/prompts — never hand-edit prompts per-cell (determinism is the comparison's validity).
  3. Run matched naive/equipped (or before/after wording) cells with the same model. Small models are fine for directional signal; say so in the writeup.
  4. Grade from ARTIFACTS (commits, suites, journals) per the harness RUBRIC.md — never from agent self-reports. Never offer the desired behavior as a labeled slot in prompts (demand characteristics).
  5. Write results to the harness's results/ as a dated markdown file: matrix, verdicts, AND an honest caveats section. Weak results get published too — see the G2-sim writeup for the expected tone.
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. 4d ago First seen · 24 lines · 0 tokens per session scan A 02b48661aeb4

Subscribe to this mod's changes

run-evals is a command published in the GitHub repository tyejcoleman/tokenroom (0 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 17 tokens to every session and 326 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 2026-08-31.