skill-forge-grader

An evaluation agent for checking whether a coding-skill output satisfies predefined requirements. It reads the evaluation rules and output files, then records pass or fail results with evidence.

In plain words
What is it for?
Use it to grade evaluation runs, verify output assertions, quote supporting evidence, and write the results to grading.json.
Why use it?
It makes skill testing more consistent by checking every assertion and calculating an overall pass rate.

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/agricidaniel/skill-forge/skill-forge-grader
Clone the repo
git clone --depth 1 https://github.com/AgriciDaniel/skill-forge
Per session 64 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 458 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 $0.00064 $0.00458
Opus 5 $0.00032 $0.00229
Sonnet 5 $0.00013 $0.00092
Haiku 4.5 $0.00006 $0.00046

Measured 2d ago against content hash b5f876c4e789, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

skill-forge-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/skill-forge-grader.md · 65 lines

What it actually says

You are an eval grading specialist for Claude Code skills.

Your Role

Evaluate skill outputs against the assertions defined in eval_metadata.json. For each assertion, determine pass/fail and provide specific evidence from the outputs that supports your judgement.

Process

  1. Read eval_metadata.json from the eval directory for assertions
  2. Read all files in the outputs/ directory of the run
  3. For each assertion: a. Check the output content against the assertion's check description b. Determine pass (true) or fail (false) c. Quote specific evidence from the output (line numbers, text excerpts)
  4. Calculate overall pass_rate: passed_assertions / total_assertions
  5. Write grading.json to the run directory

Grading Rules

  • Be strict: the output must clearly satisfy the assertion
  • Partial credit is not supported — each assertion is binary pass/fail
  • If the output is empty or missing, all assertions fail
  • If an assertion is ambiguous, grade it as fail and note why in evidence
  • Weight field in assertions is for benchmark aggregation, not grading

Output Format

Write grading.json:

{
  "eval_id": 0,
  "run_type": "with_skill",
  "assertions": [
    {
      "name": "assertion-name",
      "passed": true,
      "evidence": "Found expected output on line 14: 'Score: 87/100'"
    }
  ],
  "passed_count": 1,
  "total_count": 1,
  "pass_rate": 1.0
}

Return a summary with per-assertion results and overall pass rate.

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 · 65 lines · 64 tokens per session scan A b5f876c4e789

Subscribe to this mod's changes

skill-forge-grader is an agent published in the GitHub repository AgriciDaniel/skill-forge (166 stars, last pushed 4mo ago), licensed MIT. It adds 64 tokens to every session and 458 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 2026-08-30.

Related

Other agents, from other repositories

wiki-lint

Read-only interpreter for the deterministic portable vault linter. Runs the linter against an explicitly selected vault or scope, validates surprising findings against source pages, and returns a structured health report. It never writes reports or repairs the vault.

AgriciDaniel/claude-obsidian · 50 tokens

AGENT_AUTONOMY

Purpose: Dev agents (Theo/Nova/Pixel/Ops) autonomously source tasks from GitHub, implement them with acpxrun, and open PRs — without human direction.

Team-Commonly/commonly · 0 tokens

WEBHOOK_SDK

Write a custom Commonly agent in 30 lines of Python. The SDK is a single stdlib-only file that implements the four CAP verbs; the scaffolder wires publish + install + token-issuance in one command.

Team-Commonly/commonly · 0 tokens

AGENT_CODING_CAPABILITY

This doc exists because the answer to "why can't my OpenClaw agent just write the code?" is non-obvious and has bitten us in production. It is the source of truth for the runtime → coding-capability mapping.

Team-Commonly/commonly · 0 tokens

BUILDING_AN_AGENT

Three ways to add an agent to Commonly, from easiest to most flexible.

Team-Commonly/commonly · 0 tokens

public-facing-agent-sandboxing

If you attach an agent to a pod that strangers can read or join — a community support bot, a showcase greeter — that agent responds to untrusted input. Treat every message it receives as a potential prompt injection, and configure it so a successful injection has nothing to reach for.

Team-Commonly/commonly · 0 tokens