Skip to content

mohitagw15856/pm-claude-skills / plugins/pm-ai/skills/model-card

model-card

Verified

Document a deployed ML/AI model so others can use it responsibly. Use when asked to write a model card, document a model's intended use and limitations, or prepare an AI model for review/launch. Produces a complete model card — intended use, training data, evaluation metrics across slices, limitations, ethical considerations, and a deployment checklist.

Model Card Skill

A model card is the README for a model: what it does, what it was trained and evaluated on, where it works, and — most importantly — where it doesn't. It turns an opaque artifact into something a reviewer, a downstream team, or a regulator can actually assess. Write it before launch, not after.

Required Inputs

Ask for these only if they aren't already provided:

  • Model name & version, owner team, and date.
  • What it does — task type (classification, generation, ranking, extraction…) and the decision it informs.
  • Intended use & users — the supported use cases, and explicitly the out-of-scope ones.
  • Training data — sources, size, time range, and known gaps (link a dataset-datasheet if one exists).
  • Evaluation — datasets, metrics, and results, ideally broken down by subgroup/slice.
  • Known limitations & risks — failure modes, bias findings, safety concerns.

Output Format

Model Card: [name] v[version]

Owner: [team] · Date: [date] · Status: [in review / production / deprecated]

1. Overview — one paragraph: what the model does, the decision it serves, and who uses it.

2. Intended Use

  • In scope: the use cases this model is validated for.
  • Out of scope / do not use for: explicit prohibited or unvalidated uses (this section prevents the most harm).
  • Users: who is expected to operate or consume it.

3. Training Data — sources, size, time window, labelling method, and known coverage gaps.

4. Evaluation

  • Metrics: the primary metric(s) and why they were chosen for this task.
  • Overall results: headline numbers vs. a stated baseline.
  • Sliced results: a table of the key metric across important subgroups (geography, language, device, demographic where appropriate) — surface where performance drops, don't hide it behind an average.
SliceNMetricvs. overall

5. Limitations & Failure Modes — concrete situations where it underperforms or should not be trusted.

6. Ethical Considerations & Bias — fairness findings, sensitive-attribute handling, and mitigations applied.

7. Deployment & Monitoring — serving constraints (latency/cost), the drift/quality signals you'll watch, and the rollback trigger.

Quality Checks

  • "Out of scope / do not use for" is filled in with specifics — not left blank
  • Evaluation is reported by slice, not just one global average that hides subgroup harm
  • Every metric states the baseline it's measured against
  • Limitations describe real, concrete failure situations (not "the model may be imperfect")
  • A monitoring signal and an explicit rollback trigger are named

Anti-Patterns

  • Do not report a single aggregate metric and call evaluation done — averages mask the slices where a model fails worst
  • Do not leave "intended use" open-ended — an undefined boundary is an invitation to misuse
  • Do not omit known biases because they're uncomfortable — an undocumented risk is a worse liability than a documented one
  • Do not present accuracy without the class balance / base rate — 95% accuracy on a 95/5 split is meaningless
  • Do not ship without a monitoring plan — a model card without a rollback trigger is a snapshot, not a contract

Based On

Model Cards for Model Reporting (Mitchell et al., 2019) and the model-documentation practice used in responsible-AI reviews.

Example Trigger Phrases

  • "Write a model card."
  • "Document a model's intended use and limitations."
  • "Prepare an AI model for review/launch."

Run this skill in your browser: https://mohitagw15856.github.io/pm-claude-skills/skill/model-card.html

writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 322kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 118kUpdated Claude CodeCodexVerifiedInstall