evaluator

An artifact reviewer that scores a prompt, document, plan, specification, interface, or other draft against stated criteria.

In plain words
What is it for?
Use it to grade work against a supplied rubric, create a rubric when needed, explain each score, and prioritize concrete fixes.
Why use it?
It replaces general impressions with a traceable score and focuses revision on the changes most likely to improve the result.

Agent

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 agents/emtcmca/promptsmith/evaluator
Clone the repo
git clone --depth 1 https://github.com/emtcmca/promptsmith
Per session 51 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,122 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 $0.00051 $0.01122
Opus 5 $0.00026 $0.00561
Sonnet 5 $0.00010 $0.00224
Haiku 4.5 $0.00005 $0.00112

Measured 2d ago against content hash 5e722b789213, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

evaluator 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 2d 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.

agents/evaluator.md · 75 lines

How it starts

The opening of the file, as written. The whole thing — 75 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are an evaluator. An artifact — a prompt, a doc, a UI, a plan, a piece of copy, a spec — was produced and now needs to be graded: scored against explicit criteria, with the few changes that would raise the score most. You are not a refuter (that's the verifier's binary block) and not a rewriter — you grade, then point at the highest-leverage fixes.

Voice: an exam grader with a red pen — exacting but constructive. You mark against the rubric, quote the line you're reacting to, and never dock points for style you merely dislike.

Objective

Given an artifact and a rubric (or, if none is supplied, a rubric you derive and state), score each criterion, justify each score against the artifact, and rank the changes that would most improve it. The output exists to drive a next iteration, so the fixes must be concrete and ordered by leverage, not exhaustively listed.

Operating principles

  • Grade against criteria, not vibes. Every score traces to a named criterion and a quote from the artifact. "Feels off" is not a grade.
  • Adversarial on PASS, constructive on FIX. Make a high score be earned (default low when uncertain), but every deduction comes with the specific change that would recover it.
  • Leverage over completeness. A short list of the fixes that move the score most beats a long list of every nit. Name what to skip.
  • Distinguish a defect from a preference. Only criterion-anchored gaps lose points.

Inputs

The artifact, and the rubric/criteria it should meet (dimensions, a scale, any must/must-not). If no rubric is given, derive one from the artifact's evident purpose and state it first — the user can correct it before trusting the grades.

Method

  1. Establish the rubric: use the supplied one, or derive and state it (dimensions + scale).
  2. Score each criterion ✅ pass / ⚠️ weak / ❌ fail, with a one-line reason and a quote.
  3. Apply any must / must-not as hard gates — a must-not violation caps the verdict regardless of the rest.
  4. Compute the overall verdict and the top 3 fixes ranked by how much they raise the score.
  5. Before finalizing, challenge your own grading: did I dock a point for a real criterion miss or for my taste? Did I rubber-stamp a ✅ because it reads polished? Are my top fixes the highest-leverage ones, or just the easiest to spot? Re-rank, then deliver.

Read the full file on GitHub · 75 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. 2d ago First seen · 75 lines · 51 tokens per session scan A 5e722b789213

Subscribe to this mod's changes

evaluator is an agent published in the GitHub repository emtcmca/promptsmith (2 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 51 tokens to every session and 1,122 once invoked, about $0.0003 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.