model-card

name	model-card
description	Generate a model card covering overview, intended use, evals, limitations, governance, and versioning

Skill: /model-card — Generate a Model Card

Trigger

User runs /model-card followed by a model or feature description, or runs it alone.

Behavior

1. Ask (if not provided): model name/version, use case, training/fine-tuning data summary, known limitations, and who owns this model in production
2. Generate a complete model card — do NOT leave sections blank; write "Not yet assessed" with a [TODO] tag if data is genuinely unavailable
3. Flag governance gaps as [RISK]
4. Recommend this card be stored in /decisions/ or a dedicated /model-cards/ directory

Output Format

Model Card: [Model Name vX.Y]

Date: [today]
Owner: [team / person]
Status: [Development / Staging / Production / Deprecated]

---

1. Model Overview

Field	Value
Model ID
Base model (if fine-tuned)
Model type	(e.g., LLM, classifier, embedder)
Framework
Hosting	(cloud provider, region, endpoint type)
Version
Last updated

2. Intended Use

• Primary use cases: What this model is designed to do
• Out-of-scope use cases: What it should NOT be used for (be specific)
• Target users: Who will interact with this model directly or indirectly

3. Training & Fine-Tuning Data

• Data sources and approximate size
• Date range of training data (knowledge cutoff if applicable)
• PII / sensitive data handling during training
• Data governance / consent status
• [RISK] flag if training data provenance is unclear

4. Evaluation Results

Benchmark / Eval	Score	Baseline	Date	Notes

Include both task performance metrics and safety/fairness metrics. If evals haven't been run, flag [RISK: HIGH].

5. Known Limitations & Failure Modes

Bullet list of known weaknesses, hallucination tendencies, demographic biases, or domain gaps. Be specific — vague limitations are not useful.

6. Risks & Mitigations

Risk	Severity	Mitigation in Place

7. Governance

Item	Status
PII handling policy documented	✅ / ❌
Data retention policy defined	✅ / ❌
Audit logging enabled	✅ / ❌
Human review process for high-stakes outputs	✅ / ❌
Incident response plan exists	✅ / ❌
Approved by AI governance board / review	✅ / ❌

8. Versioning & Change History

Version	Date	Change Summary	Approved By

9. Contact & Ownership

• Model owner (team + individual)
• On-call / incident contact
• Feedback channel for end users

Quality Bar

• A model card without eval results is not production-ready — flag [RISK: HIGH] and recommend /eval-design
• "Not yet assessed" is acceptable; blank fields are not
• Governance section must be complete before the model card is used for a production deployment review