Combined diagnostic that assesses knowledge graph quality in one pass.
Optional flags: $ARGUMENTS
No arguments: quick health summary (stats + top issues)
--full: deep audit (stats + weak nodes + orphans + frontier + staleness)
--fix: auto-fix found issues (prune orphans, decay stale nodes)
consolidate(action: "stats")Returns: node count, edge count, type distributions, confidence stats (mean/min/max/std_dev), orphan count.
| Metric | Healthy | Warning | Critical |
|---|---|---|---|
| Orphan ratio | < 10% | 10–30% | > 30% |
| Mean confidence | > 0.5 | 0.3–0.5 | < 0.3 |
| Edge/node ratio | > 1.5 | 0.5–1.5 | < 0.5 |
| Confidence std_dev | < 0.25 | 0.25–0.35 | > 0.35 |
Report a summary like:
Graph Health: Good — 142 nodes, 238 edges (1.68 edge/node), mean confidence 0.64, 8 orphans (5.6%)
consolidate(action: "query", operation: "list_nodes", min_confidence: 0.0, max_confidence: 0.3, limit: 20)Nodes below 0.3 confidence are speculative — candidates for reinforcement or pruning.
retrieve(action: "frontier", limit: 10)Identifies where uncertainty is highest — the most valuable areas to investigate next.
retrieve(action: "episodic", since: null, until: null, limit: 5)If the most recent episodic node is old, the graph may be stale.
If orphan count > 10%, list them:
consolidate(action: "query", operation: "list_nodes", limit: 50)Cross-reference with edge data to identify which nodes lack connections.
When --fix is passed, take corrective action on found issues:
For each orphan node:
Search for related nodes: consolidate(action: "query", operation: "similarity_search", query: "<orphan content>", limit: 3)
If similarity > 0.5 with any node → create related_to edge
If no matches and confidence < 0.3 → prune: act(action: "forget_node", node_id: "<id>", strategy: "soft")
For nodes with confidence < 0.2:
Check if still relevant (search for supporting context)
If supporting evidence exists → learn(action: "from_feedback", node_id: "<id>", feedback_type: "positive")
If no support → act(action: "forget_node", node_id: "<id>", strategy: "soft")
If graph is stale (no episodic nodes in 7+ days), recommend running /graphonomous:sync to ingest recent filesystem changes.
## Graph Health Report
**Overall: [Good|Warning|Critical]**
| Metric | Value | Status |
|-----------------|-------|--------|
| Nodes | 142 | — |
| Edges | 238 | — |
| Edge/Node Ratio | 1.68 | ✓ |
| Mean Confidence | 0.64 | ✓ |
| Orphans | 8 (5.6%) | ✓ |
| Weakest Nodes | 3 below 0.2 | ⚠ |
**Top Issues:**
1. 3 nodes below 0.2 confidence — candidates for reinforcement
2. Frontier suggests investigating: [topic areas]
**Recommendations:**
- Run `/graphonomous:sync` to update filesystem nodes
- Review weak nodes: [list IDs]| Situation | Follow-up |
|---|---|
| High orphan count | /graphonomous:sync --full to rebuild edges |
| Low mean confidence | /graphonomous:consolidate to merge and strengthen |
| Frontier gaps | /graphonomous:retrieve <frontier topic> to fill gaps |
| Stale graph | /graphonomous:watch <dir> for continuous sync |
Running health checks too frequently — once per session or after major changes is enough
Auto-fixing without reporting what was changed — always show the user what --fix did
Treating orphans as always bad — some nodes are intentionally standalone (e.g., one-off observations)
Ignoring the frontier — it tells you where to invest learning effort next