/safina:debug
Triage production and staging issues using observability data — New Relic, git history, local logs, and browser console — and produce a structured hypothesis with specific file and line pointers.
When to run it
- Something is broken in staging or production and you don’t know why
- Error rate spiked after a deploy
- A specific endpoint is slow or throwing 5xx responses
- A local dev environment is behaving unexpectedly
- You need to link an incident to a recent code change
What it pulls
| Source | What it checks |
|---|---|
| New Relic | Error rate, p95 latency, active alerts, recent error logs |
| Git history | Commits in last 48h touching the affected service or config |
| Local logs | logs/, *.log, application stdout |
| Code | Route handler, service function, and middleware in the request path |
| Chrome browser | Console errors, failed network requests (if Chrome MCP connected) |
New Relic is Ilmiya’s primary observability platform. Service names it uses: ilm, iam, ai, users, content, omni, iqra, edify, insights, console, portal, aide.
Connect New Relic at claude.ai/settings to get live APM data, traces, and logs in the debug session.
The report format
## /safina:debug — YYYY-MM-DD HH:MM
Service: [service name]
Environment: [staging / production / local]
Symptoms: [one sentence]
### Data pulled
| Source | Signal | Value |
New Relic | Error rate (30m) | [%]
New Relic | p95 latency | [ms]
Git | Last deploy | [sha · message · time ago]
### Hypothesis
Most likely cause: [specific — names the function or query, not "something broke"]
Evidence: [grounded in data above]
Where to look first: [file:line]
### Next steps
1. [Concrete first action]
2. [If inconclusive]
3. [Escalation path]
Connections required
| Tool | Why | Connect at |
|---|---|---|
| New Relic MCP | APM, traces, error logs | claude.ai/settings |
| Chrome browser MCP | Console errors, network tab (frontend issues) | claude.ai/settings |
Both are optional — the command works without them using local context only. But New Relic makes the difference between guessing and knowing.
What it does not do
- Does not modify code — surfaces the problem, you fix it
- Does not run load tests — use
/safina:loadfor that - Does not deploy or rollback — coordinate with the team for production incidents
Offer to log to ticket
After triage, the command offers to post the summary as a comment on your active Linear or GitHub issue so the debugging context is preserved alongside the work.