Context
In November 2026 I led a read-only security audit of a content-protection feature and wrote the plan to fix what it found. The feature spans four production services, and the audit was needed before a new admin console shipped.
The problem
Before shipping, nobody had checked what protection the production code actually enforced. The risk was building new admin tools on top of controls that did not work as assumed.
What I did
I audited the four services at the exact versions deployed in production, which produced about 120 findings with file and line references and a severity for each. I re-checked the critical ones by hand.
I wrote a phased remediation roadmap. Phase 0 stops active leaks, phase 1 moves secrets out and then rotates them, and phase 2 removes bypasses. Phase 3 makes structural fixes, and phase 4 lets the admin console proceed, gated on the first two.
I wrote the working rules: move before rotate, measure before enforce, and hotfix from the production version.
I wrote a test-first plan for the most urgent hotfix, including a guard against logging key material.
I also cut the admin stories down to the ones that were honest to ship.
Key decision
Move before rotate. Moving secrets out first, with old and new signing keys both accepted for a time, meant long-lived mobile sessions would not log everyone out. Rotation came after.
Measure before enforce. New blocks first run in a counting mode, so I could see what would be blocked before blocking real users.
Result
The audit delivered about 120 findings and a roadmap in five phases, and the roadmap became the gate for the admin work. Remediation is in progress. I do not report how many findings are fixed.
What I learned
I would again audit the code that is deployed and not the code on the main branch. Next I would track fixes in the roadmap itself, so progress is visible without a separate count.
Stack
- Ruby
- Node
- C/C++
- TypeScript
