| name | owl-diagnostics |
| author | liurui |
| description | Use the owl CLI for Guance queries, diagnostics, root-cause analysis, and structured Markdown reports. |
Owl Diagnostics Skill
Use owl to gather Guance evidence, classify issues, diagnose likely causes, and write concise Markdown reports.
Core Rules
- Run
owl -h before using the tool in a new environment.
- Run
owl show <tool> before calling a specific tool; do not guess parameters.
- Prefer read-only queries unless the user explicitly asks for a write action.
- When DQL is involved, validate the DQL before querying.
- If a tool returns a file path, read the file and inspect its internal
success or error fields.
- A shell exit code of
0 does not by itself prove that the query succeeded.
Tool Selection
- Errors:
owl.errors.list.
- Logs and complex DQL:
owl.data.query.
- APM context:
owl.apm.list.
- Events:
owl.event.list.
- Metrics:
owl.metric.list.
- Network:
owl.network.list.
For complex or cross-domain questions, start with discovery, validate DQL, query the closest data domain, then add supporting evidence from logs, metrics, events, or APM.
Report Requirements
Every delivered diagnostic result should include:
- Absolute time range.
- Data domain and tools used.
- Overall conclusion.
- Evidence summary.
- Inferences clearly labeled as inferences.
- Next steps and remaining gaps.
Save final reports under ./owl-reports/ when the task asks for a report or when query results are part of the final deliverable.
Useful Commands
owl -h
owl show owl.errors.list
owl exec owl.errors.list --start_time <START_MS> --end_time <END_MS> --page_size 100
cat /tmp/owl-report.md | python3 scripts/save_report.py --output-dir ./owl-reports