Real checkups make the scoring model better, but raw reports are too sensitive for public GitHub issues. Use the contributed-report format when you want to share a grade without sharing paths, secrets, customer data, or full local findings.
python3 scripts/validate_contributed_report.py samples/contributed-report.jsonThe validator accepts only aggregate fields and fails closed when it sees raw user paths, emails, API-key shapes, bearer tokens, or other secret-shaped text.
This is an abridged example. See
samples/contributed-report.json for a
complete valid file with all 10 domain summaries.
{
"schema": "claude-code-doctor-contributed-report-v1",
"environment": {
"os": "macOS",
"harness": "Claude Code",
"doctor_version": "v0.11.0",
"audit_date": "2026-07-02"
},
"overall": {"grade": "B", "score": 72},
"metrics": {
"always_on_tokens": 18400,
"permissions": 220,
"mcp_tools": 31,
"critical_findings": 0,
"total_findings": 42
},
"domains": [
{"name": "Directory structure", "grade": "B", "score": 74, "finding_count": 4, "red_flag_count": 0}
],
"feedback": {
"grade_felt": "about_right",
"notes": "Aggregate feedback only."
}
}Rules:
- Include exactly 10 domain summaries.
- Use grades
AthroughEand scores from0to100. - Use non-negative integer counts for all metrics.
- Keep notes general: no raw paths, emails, secrets, patient data, customer data, or full report excerpts.
- Overall grade and score.
- OS family and harness.
- Aggregate metrics: token load, permission count, MCP tool count, critical finding count, total finding count.
- Per-domain grade, score, finding count, and red-flag count.
- A short note about whether the grade felt too harsh, too soft, or about right.
- Full local reports.
treeoutput with real directories.- Raw action prompts containing local paths.
- Secret-looking strings, emails, tokens, account IDs, patient data, or customer data.