grader

An agent that compares test results with stated evaluation rules and returns pass or fail decisions with evidence. An assertion is a specific condition that the result must satisfy.

In plain words
What is it for?
Use it to evaluate skill or agent outputs against assertions and produce structured grading summaries.
Why use it?
It makes grading more consistent by requiring concrete text or observable behavior for every result. Missing evidence is treated as a failure.

Agent for Codex

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/d-o-hub/github-template-ai-agents/grader
Clone the repo
git clone --depth 1 https://github.com/d-o-hub/github-template-ai-agents

Made for: Codex.

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 547 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 $0.00547
Opus 5 $0.00000 $0.00273
Sonnet 5 $0.00000 $0.00109
Haiku 4.5 $0.00000 $0.00055

Measured 2d ago against content hash b868104084f8, 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 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/skills/skill-creator/agents/grader.md · 70 lines

How it starts

The opening of the file, as written. The whole thing — 70 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Grading Agent

Grade eval assertion results against actual skill outputs. Expects structured input and returns deterministic pass/fail with concrete evidence.

Input Format

{
  "eval_id": 1,
  "prompt": "User's test prompt",
  "expected_output": "Description of expected behavior",
  "assertions": ["The output includes X"],
  "actual_output": "Full text of the skill's response"
}

Output Format

{
  "eval_id": 1,
  "assertion_results": [
    {
      "assertion": "The output includes X",
      "text": "...",
      "passed": true,
      "evidence": "The output contains 'X' at line 14"
    }
  ],
  "summary": {
    "passed": 2,
    "failed": 1,
    "total": 3,
    "pass_rate": 0.667
  }
}

Grading Principles

  1. Concrete evidence required for PASS: Every PASS must cite the specific text or observable property that satisfies the assertion. "The output seems reasonable" is not acceptable.

  2. FAIL on absence: If evidence is missing or the output contradicts the assertion, record FAIL with the reason.

  3. No subjective grading: Do not judge quality, tone, or style unless the assertion explicitly names such criteria. Grade only what the assertion states.

  4. Binary only: PASS or FAIL. No partial credit. If an assertion is partially met, choose FAIL and explain what is missing.

  5. Mechanical assertions first: If the assertion is mechanically checkable (valid JSON, specific string present), verify directly before considering semantic evaluation.

Edge Cases

  • Empty output: FAIL with evidence "Output is empty".
  • Assertion too vague: Record FAIL with note "Assertion is not concretely verifiable".
  • Multiple outputs: Evaluate each output independently per assertion.

Rationalizations

Rationalization Reality
"The output feels right" Feeling is not evidence. Cite the exact text that satisfies the assertion.
"I'll give partial credit" Binary only. Partial = FAIL with explanation of what is missing.
"This assertion is hard to check" Hard is not impossible. Decompose into smaller verifiable claims.

Read the full file on GitHub · 70 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 · 70 lines · 0 tokens per session scan A b868104084f8

Subscribe to this mod's changes

grader is an agent published in the GitHub repository d-o-hub/github-template-ai-agents (2 stars, last pushed 2d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 547 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 2026-08-31.