| name | memory-search |
| description | Search OpenSRE memory for past investigations similar to what you are seeing now. Use after you have concrete evidence (error message, failing component, stack trace)—not on vague initial alerts alone. |
| allowed-tools | Bash(python *) |
memory-search
Search OpenSRE memory for past investigations similar to what you are seeing now.
When to use
Use memory search after you have concrete evidence—not on vague initial alerts alone:
- You have an error message, stack trace, or failing job output
- You identified a failing component or service
- A symptom may have been investigated before and you want prior root causes or skills that worked
Do not search on ticket titles alone (e.g. "build pipeline failing"). Gather facts first (read Jenkins console, check logs), then search with an enriched query.
Query guide
Build queries from symptom + component + system:
- Symptom — what went wrong (OOM, connection pool exhausted, 503, test failure)
- Component — where it failed (surveys-module, checkout, payment-db)
- System — context (Jenkins, Kubernetes, PostgreSQL)
Usage
python .claude/skills/memory-search/scripts/search.py \
--query "Maven OOM surveys-module test failure Jenkins" --limit 5
JSON output
Returns JSON to stdout:
{
"success": true,
"result": {
"episodes": [...],
"strategy": "..."
}
}
episodes
Each match includes:
| Field | Description |
|---|
issue_type | Classified issue category |
root_cause | Root cause from the past investigation |
resolved | Whether the prior incident was resolved |
summary | Brief investigation summary |
skills_used | Skills that helped resolve it |
services | Affected services |
score | Similarity score (higher = closer match) |
Prioritize resolved episodes with matching root cause and skills_used.
strategy (when available)
When two or more strong matches exist, the result may include a strategy field with a playbook synthesized from similar past investigations. Use it alongside the individual episodes.
On failure, success is false and error describes the problem. Failures are non-blocking—investigation continues.
