Hey everyone! I’ve been building MindShield, an open-source browser extension designed to help break mindless AI prompt habits by evaluating intent 100% locally on your device.

I recently dropped the v1.1.5 update, so I wanted to share a quick devlog on how I solved a few Manifest V3 technical hurdles:

• Offscreen WASM Execution: To run local ONNX models without service worker timeouts, I routed the pipeline through Chrome’s Offscreen API to keep the model warm in memory.

• Zero Typing Lag: Intercepted prompt submissions using lightweight async regex checks before triggering local evaluation.

• Smart Classification: Refined the intent model so direct factual lookups pass through immediately without nuisance triggers.

What’s New in v1.1.5:

• Streamlined overlay UI (removed the old prompt bar).

• Refined classification logic to minimize false positives.

• Optimized local WebAssembly execution speed.

I’d love to hear any feedback on the workflow or answer technical questions about running local WASM in extensions!

MindShield v1.2.0: Changes

🚩Milestone - Step Towards Real Useability

This update makes MindShield a true useable extension, unlike the annoying extension from before.

major bug fixes

• large prompts do not have the triggering problem anymore
• Fixed issues with the timer some times breaking
• Added more protection to try and prevent any unseen issues that could happen when navigating while having a active reflection time

steps to true detection

• warning’s - before reflection timers were the only out come from flagged prompts now warnings exist to make sure gray areas are covered while making reflection timers keep to prompts that need real friction from a reflection timer

better detection

• Improved detection from tweaks to the model

full change logs