WHAT I DO

Read. Think. Fix.

Three things, in that order. Every time. Watch.

Ticket arrives
I diagnose
I ship
You're notified
1
I diagnose. Real root causes, grounded in your code.
  • Read every ticket your customer sends — the full body, not just the subject.
  • Clone your repo, grep the code, read recent commits.
  • Pull your deploy logs, error traces, query the DB read-only.
  • Form a hypothesis grounded in code, not vibes.
  • If anything's ambiguous, ask the customer specific questions. No fishing.
2
I ship. Smallest change, with a preview, behind your approval.
  • Branch off main with the smallest change that fixes the bug.
  • Run your smoke checks. If they fail, I fix and re-run — I don't ship a red build.
  • Push, open a PR, attach the preview deploy URL.
  • Email your reviewer with the root cause, PR link, and smoke results.
  • If commits land after approval, that approval expires. No silent merges.
3
I close the loop. Customer gets a plain-English summary. You get a receipt.
  • Once your reviewer approves on the current SHA, I merge to main.
  • Production deploys. I verify the original report is actually fixed.
  • Email the customer with a plain-English summary of what changed.
  • Every step logged immutably. Full audit, queryable any time.
  • Cost telemetry per ticket so you always know what an hour of me costs.
×
What I won't do. Hard limits — by design, not by config.
  • Touch authentication, payments, or PII without a named human signing off.
  • Push to main without a verified approval on the current head SHA.
  • Run destructive SQL — no DDL, DML, migrations. Reads only.
  • Rotate, transmit, or generate credentials. Ever.
  • Expand scope. Adjacent issues become new tickets.
  • Send unsolicited customer email. I reply to tickets — I don't broadcast.
  • Use --no-verify, force push, or hard reset without explicit human ask.
  • Quietly disable a test or skip a check to make a build pass.
5
The five paths a ticket takes. Every ticket lands on exactly one. No purgatory.
Branch A

Autonomous fix

In-scope, low-risk. PR → approval → merge → customer notified.

Branch B

I need your input

Vague ticket. I ask specific questions and pause. SLA timers prevent dark holes.

Branch C

No-op

Test, "site looks fine," never mind. Friendly ack, close, no noise.

Branch D

Off-limits

Auth/PII/payments. I explain why + suggest the safer human-led path.

Branch E

Stuck

Investigated, can't safely fix. I escalate with what I tried.

I speak fluent infrastructure.

If your stack has a sane API, I can probably read from it. Writing? Only when you say so.

Databases

Read-only by default

Supabase PostgreSQL MySQL MongoDB DynamoDB
Cloud & hosting

Deploy logs & build status

Vercel Netlify AWS Azure GCP
Observability

Errors, traces, sessions

Sentry DataDog LogRocket PostHog
Email & messaging

Per-tenant providers

Resend SendGrid Twilio Slack

“Sounds about right. Get me set up.”

Request access →