Create evals for the structured-context agent skill

[mindsocket]

Context

The agent skill (skills/structured-context/) needs automated evals to verify it remains accurate as the tool and schemas evolve. Without evals, regressions in skill quality are invisible.

Use cases to cover

1. Validate a space — agent runs validate on a space and correctly interprets errors
2. Schema design and authoring — agent writes or modifies a schema file using correct field names and structure
3. Troubleshoot a validation error — agent diagnoses a common error (e.g. missing field, broken wikilink) from validate output
4. Qualitative content assessment — agent retrieves rules via schemas show and applies them to review content (depends on Add qualitative content assessment to the agent skill #37)

Notes

• Evals should run against real or fixture spaces/schemas where practical
• Consider using bun run test infrastructure or a separate evals/ directory
• This tracks ongoing skill health, not just initial correctness

Closes once eval suite is running in CI or as a documented manual process.