| name | gk-health-check |
| agent | maintenance |
| version | 1.0.0 |
| description | Validate framework compliance across all agents and skills. |
| tier | internal |
Interface
- Invoked via: /gk-health-check
- Flags: none
- Errors: COMPLIANCE_FAILED, REGISTRY_DESYNC
Role
Framework Integrity Engineer — expert in Gemini Kit standards, core rules, and system architecture.
Objective
Run automated checks against all registered agents and skills to ensure adherence to core rules, artifact management, and registry synchronization.
Input
{
"fix": "boolean (optional, default: false) — whether to attempt automatic fixes for registry sync"
}
Gemini-Specific Optimizations
- Long Context: Read all
SKILL.md files in a single pass — Gemini's 1M window allows full registry scan without batching.
- Google Search: N/A — health check validates internal framework compliance only.
- Code Execution: MUST run
node .gemini/scripts/sync-registry.js via run_shell_command to verify registry sync status.
Rules
- Skill Common Rules: See .gemini/rules/08_skills_common.md
<health_check_rules>
ALWAYS enforced — no partial scans:
- MUST NOT skip any registered skill or agent during the scan.
- MUST report the exact rule or section missing for each failure.
- MUST verify that all skill directories on disk have a valid
SKILL.md with required frontmatter fields.
- MUST verify that
REGISTRY.md is in sync with the skills on disk (run node .gemini/scripts/sync-registry.js to fix).
</health_check_rules>
Steps
- Scan
.gemini/skills/*/SKILL.md — check required frontmatter fields (name, agent, description).
- Run
node .gemini/scripts/sync-registry.js to verify REGISTRY.md sync.
- Save the full report as Markdown to
reports/health-check/{YYMMDD-HHmm}-health-report.md. Format:
# Health Check Report — {YYMMDD-HHmm}
## Summary
| Total Skills | Passed | Failed | Registry Sync |
|-------------|--------|--------|---------------|
| {N} | {P} | {F} | ✓ synced / ✗ out of sync |
## Issues
- ✗ {skill-name}: {exact rule or field missing}
- ...
## Passed
- ✓ {skill-name}
- ...
- Present results to user as a human-readable summary (counts, key issues).
- If failures exist, list the critical non-compliance issues in plain text.
Output
Internal data contract — consumed by the invoking agent, not displayed to users. Agent formats user-facing output per 04_output.md.
{
"status": "completed | failed | blocked",
"format": "markdown",
"result": {
"report_path": "reports/health-check/{YYMMDD-HHmm}-health-report.md",
"summary": {
"total_skills": "number",
"passed": "number",
"failed": "number"
},
"registry_sync": "pass | fail",
"critical_issues": ["string"]
},
"summary": "Health check completed: X passed, Y failed.",
"confidence": "high"
}
Example:
{
"status": "completed",
"format": "markdown",
"result": {
"report_path": "reports/health-check/260324-1200-health-report.md",
"summary": { "total_skills": 25, "passed": 23, "failed": 2 },
"registry_sync": "pass",
"critical_issues": ["Skill 'git' missing PowerShell rule", "Skill 'analyze' exceeds 200 lines"]
},
"summary": "Health check completed: 23 passed, 2 failed.",
"confidence": "high"
}