We experimented with making generated skill bundles more agent-oriented by adding:
- required binary metadata
- fallback descriptions when no explicit description exists
- safety hints for mutating/destructive commands
- generated
CONTEXT.md from CLI-level rules
The goal was to make generated skill/context artifacts more useful for agents without requiring a separate hand-maintained layer.
Reference patch:
Would you be interested in upstreaming richer generated skill/context metadata for agent consumers?