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/cosmix/loom/loom-model-evaluationnpx skills add cosmix/loom --skill loom-model-evaluationgit clone --depth 1 https://github.com/cosmix/loomWrote 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/cosmix/loom/loom-model-evaluation)<a href="https://agentmods.dev/skills/cosmix/loom/loom-model-evaluation"><img src="https://agentmods.dev/badge/skills/cosmix/loom/loom-model-evaluation.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 | $0.00017 | $0.03763 |
| Opus 5 | $0.00009 | $0.01881 |
| Sonnet 5 | $0.00003 | $0.00753 |
| Haiku 4.5 | $0.00002 | $0.00376 |
Grade A, and why
loom-model-evaluation 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 yesterday.
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 — 265 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Model Evaluation
Overview
Choose metrics that match the business cost and data distribution, prevent leakage, validate with the right CV scheme, calibrate and threshold deliberately, and monitor for drift in production. This skill is the decision layer above sklearn/eval tooling.
Metric selection (the highest-leverage decision)
Wrong metric = confidently shipping a bad model. Pick from the cost structure and class balance, not habit.
| Situation | Use | Avoid / why |
|---|---|---|
| Rare positives (fraud, disease, churn) | PR-AUC, F-beta, MCC, recall@fixed-precision | Accuracy (a 99%-negative dataset scores 99% by predicting all-negative). ROC-AUC looks great even when precision is unusable — it ignores the huge TN base |
| FN much costlier than FP | F-beta with β>1 (recall-weighted), recall@precision floor | plain F1 (β=1 assumes equal cost) |
| FP costlier than FN | precision, F-beta β<1 | recall-optimized metrics |
| Need a probability, not a label | log loss, Brier score, calibration curve | thresholded accuracy/F1 |
| Multi-class imbalance | macro F1 (equal class weight), MCC | micro/weighted (dominated by majority class) |
| Regression with outliers | MAE, median AE, Huber | MSE/RMSE (squares dominated by outliers) |
| Regression, relative error matters | MAPE / SMAPE | RMSE; ⚠ MAPE explodes near zero and is asymmetric (penalizes over-prediction less) |
| Ranking / retrieval | NDCG, MAP, MRR | accuracy |
- ROC-AUC vs PR-AUC: ROC-AUC is invariant to class balance (baseline 0.5 always) — misleadingly optimistic when positives are rare. PR-AUC's baseline = positive prevalence, so it exposes the hard problem. Report PR-AUC for imbalanced detection.
average=trap (sklearn):weighted/microhide minority-class failure; usemacro(or per-class) when minority classes matter. Defaultbinaryassumes label1is positive.- MCC (Matthews correlation) is the most robust single scalar for imbalanced binary — high only when all four confusion cells are good.
- Always report a baseline (majority-class, random-stratified, or last-value for time series). A metric without a baseline is uninterpretable.
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.
- yesterday Changed · -37 tokens per session a139b68dfeb4
- 5d ago First seen · 265 lines · 54 tokens per session scan A 87b43d591925
loom-model-evaluation is a skill published in the GitHub repository cosmix/loom (54 stars, last pushed today), licensed MIT. It adds 17 tokens to every session and 3,763 once invoked, about $0.0001 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
voice-extractor
Capture a user's real writing voice from 5-20 prior samples, store a local voice.yaml fingerprint, and enforce it on newsjack drafts so AI tells disappear. Measures voice with named stylometry lenses (Burrows's Delta function-word vector, MATTR lexical diversity, sentence-length burstiness, Biber Dimension-1 register…
coverage-tracker
Run a Google Alerts-style keyword coverage tracker. Uses news-search for recent keyword queries, lets the LLM dedupe and classify real features versus junk, stores decisions in SQLite, and alerts only on new real coverage.
prompt-set-qa
Gate a prompt universe for schema and provenance completeness, target or campaign contamination, evidence entailment, naturalness, one-concept clarity, architecture consistency, aided status, answer leakage, and semantic duplicates. Use after realistic prompt generation and before human panel selection.
realistic-prompt-generation
Generate natural, controlled prompt variants from a target-blind design brief and prompt architecture while preserving approved jobs, acts, journeys, constraints, roles, locales, proximity bands, and evidence language. Use after architecture design and before contamination or semantic QA.
airflow
Queries, manages, and troubleshoots Apache Airflow using the af CLI. Use when working with anything related to Airflow - a DAG, a DAG run, a task log, an import or parse error, a broken DAG, or any Airflow operation. Covers listing and triggering DAGs, retrying runs, reading task logs, diagnosing failures, debugging…
delegating-to-otto
Drives Astronomer's Otto agent (astro otto) as a delegated sub-agent for Airflow, dbt, and data-engineering work. Use when the user explicitly asks to "use Otto", "ask Otto", "delegate to Otto", or "run this through Otto". Also offer Otto for Airflow 2 → 3 migrations and upgrade planning even when not named — Otto's…