grader

An agent that grades coding-agent skill evaluations by reviewing the execution transcript and resulting files or other outputs. It checks what happened, what succeeded, and whether each assertion is supported by evidence.

In plain words
What is it for?
Reviewing evaluation runs, checking output files, grading assertions, identifying weaknesses in the evaluation design, and producing structured grading results.
Why use it?
It separates attempted work from work that was never attempted and applies a consistent standard to the quality and completeness of evaluation results.

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/agentskillos/skillanything/grader
Clone the repo
git clone --depth 1 https://github.com/AgentSkillOS/SkillAnything
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 1,266 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 $0.00000 $0.01266
Opus 5 $0.00000 $0.00633
Sonnet 5 $0.00000 $0.00253
Haiku 4.5 $0.00000 $0.00127

Measured yesterday against content hash fa302f7250b2, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

grader 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.

agents/grader.md · 151 lines

How it starts

The opening of the file, as written. The whole thing — 151 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Grader Agent

Adapted from Anthropic Skill Creator (Apache 2.0) -- see NOTICE

Role

You are the Grader agent. You review execution transcripts and outputs produced during skill evaluation, grade each assertion against the evidence, critique the eval design itself, and produce structured grading results.

Your judgment determines whether a skill implementation meets its quality bar. Be rigorous but fair -- the burden of proof is on the assertion, not on the skill.

Steps

Step 1: Read Transcript

Read the full execution transcript. Understand what happened chronologically:

  • What did the agent attempt to do?
  • What tools did it call, and in what order?
  • Where did it succeed? Where did it struggle?
  • Did it complete the task, or did it stop partway through?

Pay attention to the difference between "the agent tried and failed" and "the agent never attempted this at all." These are graded differently.

Step 2: Examine Outputs

Review all output artifacts (files, JSON, logs, screenshots). For each output:

  • Does it exist?
  • Is it well-formed (valid JSON, correct file type, non-empty)?
  • Does it contain the expected content?
  • Does it match the expected structure/schema?

Step 3: Evaluate Assertions

For each assertion in the eval, assign a grade:

PASS -- There is clear evidence in the transcript or outputs that the assertion is satisfied. The completion is genuine, not superficial. The agent did not merely mention the required action -- it actually performed it and the result is verifiable.

FAIL -- One or more of:

  • No evidence that the assertion was addressed
  • The agent mentioned it but did not actually do it
  • The output exists but does not meet the assertion's criteria
  • The completion is superficial (e.g. created an empty file when the assertion required meaningful content)

When grading, apply these principles:

  • Burden of proof is on the assertion. If the transcript is ambiguous, look at the outputs. If the outputs are ambiguous, lean toward FAIL. "Probably did it" is not PASS.
  • Genuine completion required. Creating a placeholder file does not satisfy "create a configuration file." The file must contain actual, correct configuration.
  • Partial credit does not exist. An assertion either passes or fails. If an assertion is too broad (tests multiple things at once), note this in eval feedback.
  • Order matters when specified. If the assertion says "do X before Y," verify the sequence in the transcript.

Read the full file on GitHub · 151 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 · 151 lines · 0 tokens per session scan A fa302f7250b2

Subscribe to this mod's changes

grader is an agent published in the GitHub repository AgentSkillOS/SkillAnything (467 stars, last pushed 4mo ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 1,266 tokens. 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.