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.
npx agentmods add skills/zorost/ai-engineering-lab/error-analysis-50npx skills add zorost/AI-Engineering-Lab --skill error-analysis-50git clone --depth 1 https://github.com/zorost/AI-Engineering-LabWrote 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.
[](https://agentmods.dev/skills/zorost/ai-engineering-lab/error-analysis-50)<a href="https://agentmods.dev/skills/zorost/ai-engineering-lab/error-analysis-50"><img src="https://agentmods.dev/badge/skills/zorost/ai-engineering-lab/error-analysis-50.svg" alt="Measured on agentmods" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00046 | $0.01032 |
| Opus 5 | $0.00023 | $0.00516 |
| Sonnet 5 | $0.00009 | $0.00206 |
| Haiku 4.5 | $0.00005 | $0.00103 |
Grade A, and why
error-analysis-50 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 5d 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.
How it starts
The opening of the file, as written. The whole thing — 90 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Error Analysis 50
1 · Purpose
Replace guessing at fixes with a counted failure distribution, so engineering time lands on the failure class that actually costs the most.
2 · When to use
- When an eval score stalls and the next lever is unclear.
- Before any "improve quality" effort, prompt or model or retrieval.
- After any incident: the 50 reads turn an anecdote into a distribution.
3 · Inputs
- A scored system (
eval-first-development) and its per-case results file. - 50 failing or borderline cases. Fewer if the system is young, but never zero.
- A spreadsheet or table for the count.
4 · Procedure
- Pull the 50 most recent failures (or lowest-scoring cases). Recent, not cherry-picked, you want the distribution as it is, not as you remember it.
- Read each one fully: input, expected, actual, and the intermediate state (retrieved chunks, tool calls) where they exist. Do not skim.
- Assign each failure exactly one class. Invent classes as you go; keep a class only when a second case joins it. Common starters: wrong-source retrieval, format drift, hallucinated field, truncation, instruction-ignored, tool mis-call, ambiguous-input (the user's fault, not the model's).
- Count. The table, class × count, is the deliverable. Sort descending.
- Fix the largest class only. Write the hypothesis: "fixing chunk splits at tables removes class A (14 of 50)." Resist fixing two classes at once, you will not be able to attribute the movement.
- Add 2 to 4 representative cases of that class to the golden set with a commit message naming the class. The class can now never silently return.
- Re-run the eval. Record the delta and which class moved. If the class did not shrink, the hypothesis was wrong, revert and take the next class.
- Repeat until the gate clears or the top class is
ambiguous-input, at which point the spec, not the system, needs work. Say so, with the table as evidence.
5 · Anti-rationalization
| Excuse | Answer |
|---|---|
| "I already know what the failures are." | You know the failures you remember. The count remembers all of them. |
| "Reading 50 outputs is a waste of a day." | It is the cheapest day in AI engineering. Teams burn weeks fixing the wrong class. |
| "The failures are all unique." | Unclassifiable usually means un-read. Read ten more; classes emerge. |
| "Let's fix the three small classes first, quick wins." | Three classes of 2 cases each are 6 cases. The top class is 14. Do the math. |
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.
- 5d ago First seen · 90 lines · 46 tokens per session scan A a9fdeb246c8e
error-analysis-50 is a skill published in the GitHub repository zorost/AI-Engineering-Lab (309 stars, last pushed 18d ago), licensed MIT. It adds 46 tokens to every session and 1,032 once invoked, about $0.0002 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.
Other skills, from other repositories
fine-tuning-os
Use this skill when conducting a professional LLM fine-tuning engagement in Zero-Data mode via the MCP server fine-tuning-os. Triggers on: prestation de fine-tuning LLM, livraison de modèle affiné, pipeline Zero-Data, orchestration des 64 outils fine-tuning-os, cycle de vie entraînement (préparation / données…
ai-engineering-toolkit
6 production-ready AI engineering workflows: prompt evaluation (8-dimension scoring), context budget planning, RAG pipeline design, agent security audit (65-point checklist), eval harness building, and product sense coaching.
llm-ops
LLM Operations -- RAG, embeddings, vector databases, fine-tuning, prompt engineering avancado, custos de LLM, evals de qualidade e arquiteturas de IA para producao.
llama-factory
Expert guidance for fine-tuning LLMs with LLaMA-Factory - WebUI no-code, 100+ models, 2/3/4/5/6/8-bit QLoRA, multimodal support.
tinker-training-cost
Calculates training costs for Tinker fine-tuning jobs. Use when estimating costs for Tinker LLM training, counting tokens in datasets, or comparing Tinker model training prices. Tokenizes datasets using the correct model tokenizer and provides accurate cost estimates.
transformers
This skill should be used when working with pre-trained transformer models for natural language processing, computer vision, audio, or multimodal tasks. Use for text generation, classification, question answering, translation, summarization, image classification, object detection, speech recognition, and fine-tuning…