Tony garden-health

This skill should be used when checking system health, monitoring Garden status, verifying design principles, diagnosing agent issues, or running health checks. Performs periodic health checks across all 8 key design principles and reports issues to the Gardener.

install

source · Clone the upstream repo

git clone https://github.com/jaydeland/Tony

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/jaydeland/Tony "$T" && mkdir -p ~/.claude/skills && cp -r "$T/.claude/skills/garden-health" ~/.claude/skills/jaydeland-tony-garden-health && rm -rf "$T"

manifest: .claude/skills/garden-health/skill.md

Garden Health Check Skill

Synopsis

Performs comprehensive health checks across all 8 key design principles of the Gardener system and reports issues as input signals to the Gardener for action.

Trigger

"Check system health"
"Run health check"
"Monitor garden status"
"Diagnose agent issues"
"Verify design principles"
"Is the system working?"
Boot sequence (step_0)
Periodic checks (every 5 minutes via cron)

Health Checks by Design Principle

Principle	What to Check	Severity	Implementation
1. One Gardener Per Repo	seed.md exists, valid YAML, repo path configured	WARNING	Read seed.md, validate frontmatter
2. Skills Become Agents	`.claude/skills/` exists, skill.md files valid, YAML frontmatter parseable	WARNING	Scan skills dir, validate each skill.md
3. DuckDB for Everything	DB file exists, all tables present, query latency acceptable, disk space OK	CRITICAL	Connect to DuckDB, run schema check, timing test
4. GitHub SDLC	gh CLI available, auth configured, repo accessible	WARNING	`gh auth status` , `gh repo view`
5. Dashboard Visibility	Dashboard API responds, WebSocket connects, endpoints return data	WARNING	HTTP GET /api/stats, /health, WS connect test
6. Self-Learning	memory table has learnings, teach-agent skill exists	WARNING	Query memory table, check teach-agent skill
7. Teach-Agent	teach-agent skill.md exists, can invoke	WARNING	Check skill file, test invocation
8. Gardener Orchestrates	TheGardener loop running, input sources polling, agents responding	WARNING	Check core.py loop status, input source connectivity

Additional System Checks

Component	Check	Severity
Agent Sessions	Stuck sessions (running > 1hr without activity), error_count threshold	WARNING
Task Metrics	Failure rate spike, duration anomalies, cost spikes	WARNING
Tool Usage	Error rate per tool, timeout patterns	WARNING
Cost Tracking	Daily cost vs `max_cost_per_day` threshold	CRITICAL if exceeded
Memory Growth	Row counts growth, expiration cleanup	WARNING

Implementation

Run the health check script:

python .claude/skills/garden-health/scripts/health_check.py

The script outputs structured JSON:

{
  "status": "ok|warning|critical",
  "timestamp": "2024-01-01T00:00:00Z",
  "checks": [
    {
      "name": "duckdb_database",
      "status": "ok|warning|critical",
      "message": "Database connected, all tables present",
      "latency_ms": 5
    }
  ],
  "issues": [],
  "auto_fixes": []
}

Auto-Fix Behavior

Simple issues are auto-fixed:

Stuck sessions → terminated after 1hr inactivity
Expired memory → cleaned up
Connection timeouts → retry with backoff
Cache stale → flush and rebuild

Complex issues reported to Gardener:

Missing seed.md
Invalid skill frontmatter
DuckDB corruption
Cost threshold exceeded

Examples

# Full health check
python .claude/skills/garden-health/scripts/health_check.py --full

# Quick check (critical only)
python .claude/skills/garden-health/scripts/health_check.py --quick

# Check specific principle
python .claude/skills/garden-health/scripts/health_check.py --principle duckdb

Related Skills

```
teach-agent
```
- Creates new skills for unknown tasks
```
db
```
- DuckDB database operations for diagnostics
```
thegardener
```
- Receives health issues as input signals