genesis-evals

genesis-evals is a skill for Claude Code, Codex from danielmeppiel/genesis. It costs 195 tokens per session (2,561 once invoked), scanned A, original, Apache-2.0.

A maintainer-only skill for running Genesis evaluation tests against an AI model. An evaluation suite is a set of repeatable tests used to check whether changes produce the expected results.

In plain words
What is it for?
Use it to test Genesis pull requests and changes to its catalogue or core skill files, then compare results for consistency.
Why use it?
It helps validate changes to Genesis without shipping the test scenarios or maintainer tooling to normal users.

Skill for Claude CodeCodex

Written for no agent in particular: nothing here depends on one. Also seen: mentions subagents.

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/danielmeppiel/genesis/genesis-evals
Any agent
npx skills add danielmeppiel/genesis --skill genesis-evals
Clone the repo
git clone --depth 1 https://github.com/danielmeppiel/genesis

Made for: Claude Code, Codex.

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 genesis-evals

README.md
[![agentmods](https://agentmods.dev/badge/skills/danielmeppiel/genesis/genesis-evals.svg)](https://agentmods.dev/skills/danielmeppiel/genesis/genesis-evals)
Your own site
<a href="https://agentmods.dev/skills/danielmeppiel/genesis/genesis-evals"><img src="https://agentmods.dev/badge/skills/danielmeppiel/genesis/genesis-evals.svg" alt="Measured on agentmods" height="20"></a>
Per session 195 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,561 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 5.1 $0.00195 $0.02561
Opus 5 $0.00097 $0.01281
Sonnet 5 $0.00039 $0.00512
Haiku 4.5 $0.00019 $0.00256

Measured 6d ago against content hash c5fa0d887470, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade A, and why

genesis-evals 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 6d ago.

The scan reads SKILL.md. This mod also ships 3 executable files (scripts/score_run.py, scripts/spawn_record.py, scripts/validate_scenarios.py), 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.

dev/skills/genesis-evals/SKILL.md · 259 lines

How it starts

The opening of the file, as written. The whole thing — 259 lines — stays where its author put it; the contents beside it link to each section on GitHub.

genesis-evals: maintainer-side eval runner

Run the genesis self-eval suite. Steers the parent LLM session to orchestrate cold sub-agent spawns, capture responses, score deterministically, and report convergence.

Why this lives outside .apm/

Genesis ships to USERS via npx / apm install. Eval scenarios LOOK LIKE real user requests (that is the point). Colocating them under skills/genesis/evals/ would risk DISPATCH CONTAMINATION (an over-eager harness loader pulling scenario prompts into the active context) and PAYLOAD BLOAT for users who never run evals.

We also keep this OUTSIDE .apm/ because APM treats .apm/ as the publishable source root: its local-content scanner picks up anything under .apm/skills/ regardless of dev-marker, so apm pack --format plugin would leak this maintainer-only skill into the shipped artifact. Living under dev/skills/ keeps it scanner-invisible while still letting apm install --dev deploy it via the local-path devDependency in the root apm.yml.

This is the inverse of PHANTOM DEPENDENCY (referenced-but-not-bundled): BUNDLE LEAKAGE (bundled-but-not-consumed-at-runtime). See skills/genesis/assets/composition-substrate.md "Anti-patterns flagged at this step".

When to activate

  • Validating a genesis PR before merge
  • Any change to a file under skills/genesis/ (catalogue or SKILL.md)
  • Operator says "run evals", "regenerate eval matrix", "score on Opus"
  • Adding a new scenario (run validate first)

Hard rules

  • The model: field in every scenario YAML is REQUIRED. The runner REFUSES to spawn if missing. No silent default. The model is the single biggest variable in eval results.
  • Pre-spawn: ALWAYS call spawn_record.py to write the immutable <id>__<half>.spawn.json BEFORE invoking the harness's task tool. This is the source of truth for "what we asked for".
  • Cold spawn: each (scenario, half) is a SEPARATE task-tool call with fresh context. Never reuse a session across scenarios.
  • Determinism: scoring is python (score_run.py), not LLM-judged. Pass gates are substring matches against the schema.
  • Scenarios are FROZEN once landed. Removing a scenario requires setting retired_in: <version> (never deletion).
  • After 3 iteration loops without convergence, escalate via B10 HUMAN CHECKPOINT. Do NOT loop indefinitely.

Read the full file on GitHub · 259 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. 6d ago First seen · 259 lines · 195 tokens per session scan A c5fa0d887470

Subscribe to this mod's changes

genesis-evals is a skill published in the GitHub repository danielmeppiel/genesis (67 stars, last pushed 3mo ago), licensed Apache-2.0. It adds 195 tokens to every session and 2,561 once invoked, about $0.0010 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 skills, from other repositories

gardener

Monthly catalog hygiene audit for Claude Code plugin marketplaces. Inventories every skill and agent, scores each against a 7-dimension rubric via parallel Opus Agents, detects description collisions via Bedrock Cohere v4 embeddings with pairwise cosine, and writes a delta report vs. the prior audit. Proposes…

theagenticguy/agentic-plugins · 105 tokens

using-adept

Use the adept CLI to author AI skills once and render them into Claude Code, Cursor, Codex, Copilot, and OpenCode. Apply when installing/syncing skills, editing skill.yaml/SKILL.md, or touching .adeptability/.

itaywol/adeptability · 54 tokens

authoring-adept-agents

Write a good, portable adept agent (subagent): trigger-shaped description, one job per agent, generator/evaluator separation, explicit boundaries, restricted tools. Apply when creating or editing an agent file or running adept agent add.

itaywol/adeptability · 47 tokens

authoring-adept-loops

Compose a loop — a scheduled system that discovers work, hands it to agents, verifies with an independent evaluator, persists state, and reschedules itself. Apply when the user wants automation that runs on a timer, a triage/babysitter routine, or asks about adept loop add.

itaywol/adeptability · 60 tokens

authoring-adept-skills

Write a good, portable adept skill: pick the right activation, craft a triggering description, keep it scan-safe and within harness byte budgets. Apply when creating or editing a SKILL.md or running adept skill add.

itaywol/adeptability · 45 tokens

expertise-exchange

Team expertise billboard via adept exchange: ask teammates for expertise and stack responses. Apply when the user wants a colleague's input, mentions the exchange, or when you start using adept — sample open requests and offer to answer ones the user knows about.

itaywol/adeptability · 51 tokens