model-card

Writes the model card: purpose, limits, data, performance, validation status, risks and appropriate use.

SKILL.md

GitHub
<!-- Generated from .claude/skills/model-card/SKILL.md by tools/sync_agents.py. Edit the source, then run the script. -->

/model-card — Model Card

Category: Documentation · Primary agent: technical-documentation-specialist

Writes the model card: purpose, limits, data, performance, validation status, risks and appropriate use.

Arguments: whatever the user wrote after the command name.

Purpose

Write a model card: a short, factual document that tells a prospective user what the model is for, what it must not be used for, what data it uses, how well it performs, its validation status, its risks and limits. Every claim links to evidence. The output protects users from misusing the model.

Use it when a model is released, handed over or materially changed.

Do not use it when full technical documentation is required; use a technical note (/technical-note) alongside.

Before you start

  1. Check required inputs. For each input below record FOUND or ABSENT (never 'assumed'):
    • Model spec
    • Results
    • Validation report (if available) If a required input is ABSENT, that part is NOT ASSESSED — NO DATA: do not estimate or invent it; name what is missing and which skill produces it. If all required inputs are absent, stop and say so.
  2. Check confidentiality. Identify the data classification level of the inputs (.claude/docs/confidentiality-and-ai-use.md). Do not proceed with restricted or confidential data in a way the policy does not allow.
  3. Collaboration. Follow .claude/docs/collaboration-protocol.md: ask questions first, present options, get approval before writing files, and update production/session-state/active.md after each approved section.

Agents

  • technical-documentation-specialist — primary agent; coordinates the work and integrates results
  • model-validation-specialist — Performs independent model validation

Each agent is defined in .claude/agents/<name>.md. If your tool runs subagents, run these agents as subagents, in parallel where the steps are independent; otherwise read each agent's file and apply its instructions in turn. Subagents only write to the paths named in your instructions.

Key design decisions (present options, let the user choose)

Decision Typical options What it changes
Audience modelers, analysts, decision-makers Level of detail
Performance evidence backtests, benchmarks, validation results Credibility
Lego detail whole set, per-brick notes Modularity transparency

Workflow

  1. Collect facts from the spec, results and validation report
  2. Draft each section of docs/templates/model-card.md with the user's approval
  3. State intended uses and explicit non-uses
    • Include the design domain and known failure cases
  4. Verify claims against artifacts

Checkpoints with the user

  • After each section of the card

Common pitfalls

  • Performance claims without evidence links
  • Missing non-uses
  • Card not updated after model changes

Output

docs/models/<name>/model-card.md

The model card covers purpose, intended and non-intended uses, data, performance with evidence links, validation status, risks, limits, version and owner.

Gate / verdict

  • Intended and non-intended uses are explicit; every performance claim links to evidence
  • Version and owner are stated

Suggested next steps

  • /model-release
  • /technical-note