eval

A command for grading a skill against its predefined evaluation cases. An evaluation case is a test scenario with checks for whether the skill behaves as expected.

In plain words
What is it for?
Use it to list test cases, run each case in a temporary setup, collect outputs, apply rubric-based judgments, and produce a pass/fail report.
Why use it?
It provides a repeatable way to measure a skill's output and identify failed cases instead of relying only on manual inspection.

Command

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/lxb12123/agent-plugin-kit/eval
Clone the repo
git clone --depth 1 https://github.com/lxb12123/agent-plugin-kit
Per session 13 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 461 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 $0.00013 $0.00461
Opus 5 $0.00006 $0.00230
Sonnet 5 $0.00003 $0.00092
Haiku 4.5 $0.00001 $0.00046

Measured yesterday against content hash c15c54b3279e, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

eval 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 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.

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/eval.md · 33 lines

What it actually says

/eval — Evaluate a skill

Run a skill's bundled evals/ cases (e.g. for skills/review) to see whether it meets the bar.

Flow

  1. Determine the skill directory to evaluate, skills/<name> (ask the user, or take one already present in the current project).
  2. List the cases:
    node "${CLAUDE_PLUGIN_ROOT}/lib/cli.mjs" eval skills/<name>
    
    Outputs [{name, input, expect}, ...].
  3. For each case: set up the scenario as described by input (create a temporary fixture / git repo if needed), run the skill, and capture its output text.
  4. Collect all outputs into a single JSON { "<caseName>": "<output>" } and write it to a temp file, e.g. runs.json.
    • For cases with expect.rubric: you (as the LLM judge) decide whether the output satisfies the rubric description, written as an object { "<caseName>": { "output": "...", "rubric": true/false } }. The engine counts the deterministic assertions together with your rubric verdict toward pass/fail.
  5. Score:
    node "${CLAUDE_PLUGIN_ROOT}/lib/cli.mjs" eval skills/<name> --runs runs.json
    
    Outputs {total, passed, failed, cases:[{name, pass, failures}]}.
  6. Relay the report to the user; for each case with a non-empty failures, point out which expectation (contains / notContains / matches) was not met.

Notes

  • The deterministic assertions in expect (contains / notContains / matches regex) are scored by the engine, and pass/fail is decided by it.
  • When subjective quality (rubric) judgment is needed, you may add commentary, but it does not change the deterministic conclusion.
  • Evaluation is read-only on the user's project; it does not write.
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 · 33 lines · 13 tokens per session scan A c15c54b3279e

Subscribe to this mod's changes

eval is a command published in the GitHub repository lxb12123/agent-plugin-kit (1 stars, last pushed 2mo ago), licensed MIT. It adds 13 tokens to every session and 461 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.