llm-eng-checks

llm-eng-checks is a skill for Claude Code from jhlee0409/omni-harness-kit. It costs 181 tokens per session (2,125 once invoked), scanned A, original, MIT.

A checklist for building or strengthening features that use large language models (LLMs), the systems that generate text from prompts. It covers prompts, retrieval-augmented generation (RAG), testing, safeguards, model choices, and token costs.

In plain words
What is it for?
Use it when designing prompts, adding RAG, choosing a model or provider, building evaluations, adding guardrails, or measuring token and API costs.
Why use it?
It prevents a feature from being declared successful just because code runs: it requires a real model call and checks whether the response meets the intended contract.

Skill for Claude Code

Written for Claude Code: shipped in a Claude Code plugin.

Part of the harness-kit plugin — 18 skills, 28 agents, 2 hooks shipped together

Good fit Use it when designing prompts, adding RAG, choosing a model or provider, building evaluations, adding guardrails, or measuring token and API costs.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/jhlee0409/omni-harness-kit/llm-eng-checks
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.

Any agent
npx skills add jhlee0409/omni-harness-kit --skill llm-eng-checks
Clone the repo
git clone --depth 1 https://github.com/jhlee0409/omni-harness-kit

Made for: Claude Code.

Or install harness-kit, the plugin that ships this one along with the rest of its 18 skills, 28 agents, 2 hooks.

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 llm-eng-checks

README.md
[![agentmods](https://agentmods.dev/badge/skills/jhlee0409/omni-harness-kit/llm-eng-checks/github.svg)](https://agentmods.dev/skills/jhlee0409/omni-harness-kit/llm-eng-checks)
Your own site
<a href="https://agentmods.dev/skills/jhlee0409/omni-harness-kit/llm-eng-checks"><img src="https://agentmods.dev/badge/skills/jhlee0409/omni-harness-kit/llm-eng-checks/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for llm-eng-checks

Your own site · 80×15
<a href="https://agentmods.dev/skills/jhlee0409/omni-harness-kit/llm-eng-checks"><img src="https://agentmods.dev/badge/skills/jhlee0409/omni-harness-kit/llm-eng-checks.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 181 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,125 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
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 $0.00181 $0.02125
Opus 5 $0.00090 $0.01063
Sonnet 5 $0.00036 $0.00425
Haiku 4.5 $0.00018 $0.00213

Measured 11d ago against content hash b15375bfd82c, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-11, from the pricing page.

Security

Grade A, and why

llm-eng-checks 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 11d 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.

adapters/omp/skills/llm-eng-checks/SKILL.md · 91 lines

How it starts

The opening of the file, as written. The whole thing — 91 lines — stays where its author put it; the contents beside it link to each section on GitHub.

LLM engineering — design → round-trip → measure → verdict

The dominant failure pattern is claiming an LLM feature works from the code path alone. Code shows the prompt is assembled and the SDK is called; only a real round-trip shows the model actually returns what the contract promises. This skill is provider-agnostic — detect the repo's own client (a hosted provider SDK / a local inference server / an internal gateway) and drive that, never a vendor CLI you assume exists.

PRIME RULE — one real round-trip or it is not done

Before reporting any LLM change done you MUST show ONE real call:

  • Real prompt in (quote the actual assembled prompt, not a paraphrase).
  • Real output out (quote the model's actual response, not a mock).
  • Provider + model named (<provider>/<model>) + token counts.

Drive it with the repo's own client via bash (its Python/JS entrypoint) — read config from the repo env, never hardcode a key. A stub / fixture / synthetic response = "static OK, dynamic unverified", never "done". "200 OK" is not proof — the content must satisfy the output contract.

1. Prompt design

  • Contract, not vibes. Every prompt names: role, task, hard constraints (negative ones too), and an explicit output contract. If the output is consumed by code, the contract is a schema, not prose.
  • Few-shot earns its tokens or it is cut. Add exemplars only when a zero-shot round-trip fails the contract; measure the delta (pass-rate before/after). Never carry examples "just in case" — they inflate every request forever.
  • Structured output. Prefer provider JSON-schema / tool-calling for machine-consumed output. Then validate every response against the schema (jsonschema / pydantic / zod — the repo's own). An unvalidated json.loads() is a latent crash: quote a round-trip where the model returns malformed/extra fields and show the validator catching it.
  • Determinism where it matters. Pin temperature=0 (or the provider's floor) for extraction/classification; reserve higher temp for generative surfaces. State which and why.

Read the full file on GitHub · 91 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. 11d ago First seen · 91 lines · 181 tokens per session scan A b15375bfd82c

Subscribe to this mod's changes

llm-eng-checks is a skill published in the GitHub repository jhlee0409/omni-harness-kit (2 stars, last pushed 1mo ago), licensed MIT. It adds 181 tokens to every session and 2,125 once invoked, about $0.0009 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.