#878 Tell agents that rich-text fields support Mermaid diagrams

Companion to basecamp/bc3#13769. Basecamp now shows a fenced code block with the language mermaid as a diagram. This change adds that to the Basecamp skill so agents use it.

Changes

• skills/basecamp/SKILL.md: a paragraph in "Content and stdin" that says rich-text fields support Mermaid, when a diagram is the better choice than prose, and where the source is shown instead of the diagram (previews, emails, exports, and source that does not parse). An example posts a message with a diagram on stdin.
• skill-evals/cases/create-message-diagram.yml: asks for a message with a diagram and expects a mermaid fence in the command.

No CLI code changes. internal/richtext already writes a fence's language to <pre language="…"> and reads it back, which is the format Basecamp uses for code blocks. The screenshots on basecamp/bc3#13769 show a message posted with basecamp messages create.

Testing

• make check-skill-drift and make check-eval-patterns pass.
• go test ./internal/commands passes, including the 16 KiB skill budget.
• I did not run the model evals (make skill-eval), because they need an API key.

🤖 Generated with Claude Code