ai-forge-eval

ai-forge-eval is a skill for Claude Code, Codex from robcsaszar/ai-forge. It costs 112 tokens per session (2,754 once invoked), scanned A, original, MIT.

A behavioral testing workflow for agent skills and agents. It uses saved test prompts and assertions to check whether an artifact changes the model's behavior and to detect regressions over time.

In plain words
What is it for?
It is for creating or running repeatable evaluations of SKILL.md files and agent definitions.
Why use it?
A written rubric can look correct without proving that the skill or agent works in practice.

Skill for Claude CodeCodex

Part of the robcsaszar-ai-forge plugin — 8 skills, 3 agents shipped together

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 skills/robcsaszar/ai-forge/ai-forge-eval
Any agent
npx skills add robcsaszar/ai-forge --skill ai-forge-eval
Clone the repo
git clone --depth 1 https://github.com/robcsaszar/ai-forge

Made for: Claude Code, Codex.

Or install robcsaszar-ai-forge, the plugin that ships this one along with the rest of its 8 skills, 3 agents.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for ai-forge-eval

README.md
[![agentmods](https://agentmods.dev/badge/skills/robcsaszar/ai-forge/ai-forge-eval.svg)](https://agentmods.dev/skills/robcsaszar/ai-forge/ai-forge-eval)
Your own site
<a href="https://agentmods.dev/skills/robcsaszar/ai-forge/ai-forge-eval"><img src="https://agentmods.dev/badge/skills/robcsaszar/ai-forge/ai-forge-eval.svg" alt="Measured on agentmods" height="20"></a>
Per session 112 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,754 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 $0.00112 $0.02754
Opus 5 $0.00056 $0.01377
Sonnet 5 $0.00022 $0.00551
Haiku 4.5 $0.00011 $0.00275

Measured 3d ago against content hash 2a0434f9836e, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

ai-forge-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 3d ago.

The scan reads SKILL.md. This mod also ships 1 executable file (scripts/aggregate-benchmark.cjs), listed below but not scanned — reading those needs a real analyzer, not pattern matching.

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.

skills/ai-forge-eval/SKILL.md · 196 lines

How it starts

The opening of the file, as written. The whole thing — 196 lines — stays where its author put it; the contents beside it link to each section on GitHub.

AI Forge Eval

Behavioral validation for skills and agents. Rubric scoring (ai-forge-judge) tells you if an artifact is well-written. Eval tells you if it works.

Works for SKILL.md (skills) and .agent.md (agents). Same 5-phase flow; Phase 2 setup differs by artifact type.


Phase 0 — Load or Write the Suite

MANDATORY — READ references/eval-suite.md for the on-disk format, the trigger protocol, and the assertion-discrimination table.

Check for evals/evals.json beside the artifact.

  • Present — load it and skip to Phase 2. Add cases if coverage is thin; never silently replace existing ones, or the trend breaks.
  • Absent — author it in Phase 1 and write it to evals/evals.json before spawning anything.

A suite that lives only in this conversation cannot detect a regression next month. Persisting it is what makes the difference between an opinion and a test.


Phase 1 — Write Evals

Write 2–3 eval cases. Each eval is a realistic prompt plus 3–5 assertions.

Prefer prompts drawn from a baseline probe (ai-forge-create Phase 1b) over the artifact's own stated triggers — assertions written from the artifact can only confirm it does what it claims. Carry each capture into the scenario's baseline_failure field.

For skills: prompts that should activate the skill naturally. Assertions check skill-specific behaviors (e.g. "output includes a Phase 1 recap", "NEVER rule format has WHY and INSTEAD").

For agents: prompts covering the agent's stated scope. Assertions check observable behaviors — files created, tools called, tone constraints, scope limits (e.g. "did not modify files outside src/lib/", "opened a PR", "commit message starts with 'refactor:'").

Assertion rules:

  • Checkable: the assayer can verify from the output alone
  • Specific: "output includes 'Phase 1' header" beats "output is well-structured"
  • Falsifiable: must be possible to fail

Record evals:

Eval 1: <prompt>
Expectations:
- <assertion 1>
- <assertion 2>
- <assertion 3>

Read the full file on GitHub · 196 lines

Files

What ships with it

7 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

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. 3d ago First seen · 196 lines · 112 tokens per session scan A 2a0434f9836e

Subscribe to this mod's changes

ai-forge-eval is a skill published in the GitHub repository robcsaszar/ai-forge (0 stars, last pushed yesterday), licensed MIT. It adds 112 tokens to every session and 2,754 once invoked, about $0.0006 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.

Related

Other skills, from other repositories

iron-system

Meta-skill for reverse loop engineering that transforms an AI agent's approach to any complex task. Combines systematic deconstruction (reverse engineering) and iterative construction (loop engineering) in an EXTRACT→SYNTHESIZE→BUILD→AUDIT cycle. Includes 7 named tactical loops (FORGE, SIEGE, PRISM, RAZOR, MIRROR…

william1mufassa/IRON-SYSTEM · 235 tokens

skill-creator

Create or optimize Claude Code skills with best-practice structure, bundled resources, and prompt engineering patterns.

win4r/skill-creator · 24 tokens

dspy

Build complex AI systems with declarative programming, optimize prompts automatically, create modular RAG systems and agents with DSPy - Stanford NLP's framework for systematic LM programming.

synthetic-sciences/openscience · 35 tokens

guidance

Control LLM output with regex and grammars, guarantee valid JSON/XML/code generation, enforce structured formats, and build multi-step workflows with Guidance - Microsoft Research's constrained generation framework.

synthetic-sciences/openscience · 38 tokens

instructor

Extract structured data from LLM responses with Pydantic validation, retry failed extractions automatically, parse complex JSON with type safety, and stream partial results with Instructor - battle-tested structured output library.

synthetic-sciences/openscience · 41 tokens

outlines

Guarantee valid JSON/XML/code structure during generation, use Pydantic models for type-safe outputs, support local models (Transformers, vLLM), and maximize inference speed with Outlines - dottxt.ai's structured generation library.

synthetic-sciences/openscience · 50 tokens