Hi HN, here to show you guys my first ever completed project as a maintenance electrician with no CS background or programming experience. Using Claude as the architect and Claude Code for implementation, I managed to publish an extension. Now, that being said, I'm good at photoeyes and limit switches, and maybe a little less confident with browser extensions so far, but this is what I've come up with:
A Terms of Service scanner that flags concerning clauses from a number of categories. I wanted to emphasize privacy and keep everything 100% local with no API calls or data going anywhere non-local. Other extensions were sending data to external servers. With mine, there is no AI involved in the scan or the results.
I accomplished this by creating a 318 point regression suite and then coming up with a pretty accurate regex scan that includes negation awareness and lookback windows instead. The lack of AI comes with the added benefit of consistent results (same input = same output), but may come with some semantic pitfalls I have to work on in future updates. Regex only recognizes patterns, so companies can use clever wording to avoid matches, which may be grounds for a future update, but for now this is intended to remain simple and private.
Hi HN, here to show you guys my first ever completed project as a maintenance electrician with no CS background or programming experience. Using Claude as the architect and Claude Code for implementation, I managed to publish an extension. Now, that being said, I'm good at photoeyes and limit switches, and maybe a little less confident with browser extensions so far, but this is what I've come up with:
A Terms of Service scanner that flags concerning clauses from a number of categories. I wanted to emphasize privacy and keep everything 100% local with no API calls or data going anywhere non-local. Other extensions were sending data to external servers. With mine, there is no AI involved in the scan or the results.
I accomplished this by creating a 318 point regression suite and then coming up with a pretty accurate regex scan that includes negation awareness and lookback windows instead. The lack of AI comes with the added benefit of consistent results (same input = same output), but may come with some semantic pitfalls I have to work on in future updates. Regex only recognizes patterns, so companies can use clever wording to avoid matches, which may be grounds for a future update, but for now this is intended to remain simple and private.
As an example, I scanned Microsoft's ToS and it flagged 13 clauses (5 warning, 4 caution and 4 informational). This is the source document: https://www.microsoft.com/en-us/servicesagreement
For those who are interested in the audit results with the methodology, you can view that here: https://lensari.ca/fine-print/audits/microsoft.html
I've also run audits on Google, Apple, Amazon, Meta and many other major companies, which are published on the same site.
Worth mentioning: there is no account required, no telemetry and no servers involved.
Looking forward to everyone's feedback on my first public release, since I know there will be gaps to close before I try my next one!