<![CDATA[I find the #claudecode source code interesting.]]>I find the source code interesting.

It's not that different from codex or gemini. Shares many similarities.

Good job.

]]>https://board.circlewithadot.net/topic/8ff27c2c-3ef0-4f80-b56e-ceed230cb830/i-find-the-claudecode-source-code-interesting.RSS for NodeFri, 17 Apr 2026 20:51:11 GMTWed, 01 Apr 2026 17:59:29 GMT60<![CDATA[Reply to I find the #claudecode source code interesting. on Wed, 01 Apr 2026 18:31:09 GMT]]>I don't think this was a marketing gig leak. There is too much IP in there.

People have no idea how difficult this stuff can be.

]]>https://board.circlewithadot.net/post/https://infosec.exchange/users/windsheep/statuses/116330874132898658https://board.circlewithadot.net/post/https://infosec.exchange/users/windsheep/statuses/116330874132898658Wed, 01 Apr 2026 18:31:09 GMT<![CDATA[Reply to I find the #claudecode source code interesting. on Wed, 01 Apr 2026 18:28:45 GMT]]>There is a semantic firewall against hallucinations.

Of course... they use an agentic pattern for that.

]]>https://board.circlewithadot.net/post/https://infosec.exchange/users/windsheep/statuses/116330864684281804https://board.circlewithadot.net/post/https://infosec.exchange/users/windsheep/statuses/116330864684281804Wed, 01 Apr 2026 18:28:45 GMT<![CDATA[Reply to I find the #claudecode source code interesting. on Wed, 01 Apr 2026 18:25:45 GMT]]>They use Zod schemas against hallucinations.

They should write books. This is cool.

]]>https://board.circlewithadot.net/post/https://infosec.exchange/users/windsheep/statuses/116330852849444021https://board.circlewithadot.net/post/https://infosec.exchange/users/windsheep/statuses/116330852849444021Wed, 01 Apr 2026 18:25:45 GMT<![CDATA[Reply to I find the #claudecode source code interesting. on Wed, 01 Apr 2026 18:23:59 GMT]]>Ahh you don't stdin the code base. You segment it into pieces. It reads code, rather than copying it over. The client assumes that it cannot just send over x thousand lines and ask for inference.

Of course...

]]>https://board.circlewithadot.net/post/https://infosec.exchange/users/windsheep/statuses/116330845950434705https://board.circlewithadot.net/post/https://infosec.exchange/users/windsheep/statuses/116330845950434705Wed, 01 Apr 2026 18:23:59 GMT<![CDATA[Reply to I find the #claudecode source code interesting. on Wed, 01 Apr 2026 18:20:47 GMT]]>Context pruning, minimum complexity, and faithful reporting are key things to consider.

Alright. That makes total sense.

]]>https://board.circlewithadot.net/post/https://infosec.exchange/users/windsheep/statuses/116330833357650394https://board.circlewithadot.net/post/https://infosec.exchange/users/windsheep/statuses/116330833357650394Wed, 01 Apr 2026 18:20:47 GMT<![CDATA[Reply to I find the #claudecode source code interesting. on Wed, 01 Apr 2026 18:16:05 GMT]]>The "adversarial verifier" is key, imho.

I see it quite often, and it kicks in for complex challenges. I didn't know you could do that.

]]>https://board.circlewithadot.net/post/https://infosec.exchange/users/windsheep/statuses/116330814846405598https://board.circlewithadot.net/post/https://infosec.exchange/users/windsheep/statuses/116330814846405598Wed, 01 Apr 2026 18:16:05 GMT