Validate Schema

This skill allows the agent to verify that any data structure, file, YAML block, or agent output conforms to its defined schema or structural contract — producing a precise, actionable pass/fail report before the data is acted upon or written to disk.

Steps

1. Locate and read the schema. Find the authoritative schema definition:
   • For workspace YAML files → check .db/.schemas/ for the appropriate schema
   • For toolbox files → check the toolbox yaml_path file's structure
   • For custom structures → derive from the written spec or identity file
2. Read the target fresh. Apply Zero-Drift Audit Law — read the actual current file/data live, not from cached context.
3. Map required fields. List every required field from the schema. Note which are optional.
4. Check field presence. For each required field, verify it exists in the target.
5. Check field types. Verify data types match (string vs int, list vs scalar, etc.).
6. Check value constraints. Validate enum values, non-empty lists, valid paths, valid statuses, etc.
7. Check structural contract. Verify nesting depth, key naming conventions, and ordering where prescribed.
8. Compile violations. For every failure, record: field path, expected value/type, actual value/type, severity (blocking vs warning).
9. Determine verdict.
   • Zero violations → VALID ✅
   • Warnings only → VALID WITH WARNINGS ⚠️
   • Any blocking violation → INVALID ❌
10. Generate fix instructions. For each violation, produce a specific, executable correction.
11. Output the validation report.

Output Format

validation_report:
  target: "<file or data description>"
  schema_source: "<where the schema was found>"
  verdict: "VALID | VALID_WITH_WARNINGS | INVALID"
  violations:
    - field_path: "<e.g. content.skills[0].maturity>"
      expected: "<expected type or value>"
      actual: "<what was found>"
      severity: "blocking | warning"
      fix: "<exact correction to apply>"
  summary: "<one-line verdict summary>"

Notes

• Always validate before writing. Any automated file write should be preceded by a schema validation of the content.
• If no schema exists for a target, flag this as a gap and generate a schema proposal before proceeding.
• Validation failures on .db/.system.board.yaml or router files are critical — they corrupt agent routing.
• Pair with planning toolbox: if INVALID, generate a fix plan before retrying.