skill-card-generator

Automating NVIDIA governance skill card creation involves discovering signals from a target skill directory and its repository to assemble a deterministic governance card context, rendering a draft card from a Jinja template, and validating the submission for human review.

A deterministic governance card is generated by extracting frontmatter and repository signals from a target skill directory to seed a standardized context, which is then rendered through a Jinja template to ensure consistent compliance documentation.

Validating a rendered governance card requires running a validation script to surface VERIFY and SELECT markers, flagging specific fields that require manual reviewer action before the compliance documentation can be finalized.

Jinja2 supports rendering governance review artifacts by applying a style guide and template to skill-scoped frontmatter and repository signals, producing a deterministic draft card without exposing sensitive data or leaving unreconciled markers.

Refreshing compliance cards after updating agent skills is best handled by running a discovery workflow against the updated target skill directory to gather new signals, re-render the governance card, and validate the submission for any newly flagged items.

Automated skill card generation should not be used as a final submission without human oversight, because the workflow explicitly flags VERIFY and SELECT markers indicating governance fields that require manual reviewer action and validation.