test

A command that runs an LLM benchmark, saves detailed results, and creates a report explaining failures. An LLM is a large language model, and a benchmark is a repeatable set of tests used to measure its performance.

In plain words
What is it for?
It helps test models across all cases or selected categories, repeat attempts, rerun only failures, and generate failure-analysis reports.
Why use it?
It turns model test runs into saved evidence and analysis instead of requiring someone to inspect every result manually.

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/ecro/embedeval/test
Clone the repo
git clone --depth 1 https://github.com/Ecro/embedeval

Made for: Claude Code.

Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,544 The whole file, excluding the scripts and references it only reads on demand.
Security scan C 1 finding. 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 $0.00000 $0.01544
Opus 5 $0.00000 $0.00772
Sonnet 5 $0.00000 $0.00309
Haiku 4.5 $0.00000 $0.00154

Measured yesterday against content hash 934206c73cb1, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade C, and why

test scanned grade C with 1 finding 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 yesterday.

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.

Recursive force deletehighDestructive command

rm -rf with a variable or a broad path is one typo away from removing the wrong tree.

rm -rf results/claude-code*
.claude/commands/test.md · 169 lines

How it starts

The opening of the file, as written. The whole thing — 169 lines — stays where its author put it; the contents beside it link to each section on GitHub.

/test — Run EmbedEval Benchmark

Communication Protocol

  • Be direct and matter-of-fact. No flattery, no preamble, no "Great question!"
  • If reasoning is flawed, say so immediately with specific evidence
  • Don't fold arguments on pushback — maintain position unless new evidence is presented
  • Lead with concerns before agreement
  • When you agree, explain WHY with specific reasoning — not just validation

Run LLM benchmark, save detailed results, generate failure analysis report.

Usage

/test <model> [category] [--attempts N] [--retest-only] [--with-private]

Examples:

/test help                            # Show this help + category list
/test list                            # Show all categories with case counts
/test sonnet                          # All 179 public cases with Sonnet
/test sonnet kconfig                  # Only kconfig category
/test sonnet isr-concurrency          # Only isr-concurrency
/test sonnet security,ble             # Multiple categories (comma-separated)
/test sonnet --attempts 5             # 5 attempts per case (for pass@5)
/test sonnet kconfig --attempts 3     # Category + attempts
/test opus                            # All cases with Opus
/test haiku gpio-basic                # GPIO cases with Haiku
/test sonnet --retest-only            # Only re-run cases that changed since last test
/test haiku dma --retest-only         # Retest only changed DMA cases
/test sonnet --with-private           # Include 48 held-out private cases
/test haiku --with-private --retest-only  # Retest changed cases including private

Arguments

  • $ARGUMENTS is parsed as: <model> [category] [--attempts N] [--retest-only] [--with-private]
  • Model: passed to claude-code://<model> (e.g., sonnet, opus, haiku, claude-sonnet-4-6)
  • Category: optional, one of the 23 EmbedEval categories (comma-separated for multiple)
  • Attempts: optional, default 1
  • --retest-only: only run cases where TC changed since last test (uses test_tracker.json)
  • --with-private: include 48 held-out private cases from ../embedeval-private/cases/

Read the full file on GitHub · 169 lines

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. yesterday First seen · 169 lines · 0 tokens per session scan C 934206c73cb1

Subscribe to this mod's changes

test is a command published in the GitHub repository Ecro/embedeval (11 stars, last pushed 26d ago), licensed Apache-2.0. It costs nothing until one of its globs matches a file; then it loads 1,544 tokens. A static security scan graded it C with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.