Murooka

Requirements-to-design dev by day. Imagining a future where LLMs build from requirements, I learn by building what I need: LAISORA (VS Code extension for Claude Code) and prompts. I share both.

Japan Joined on Nov 6, 2026 https://www.az-llc.com/

Murooka

Nov 13

Claude Code Forgets My Instructions as Compactions Pile Up. One Added Prompt Stopped It: 0 Lost After 3 Compactions

#claudecode #claude #llm #promptengineering

22 min read

Want to connect with Murooka?

Create an account to connect with Murooka. You can also sign in below to proceed if you already have an account.

Create Account

Already have an account? Sign in

Murooka

Nov 12

Neither Forbidding nor Asking Reduced the Unwanted Behavior. Defining Where It's Allowed Did. Tested on Claude Code

#claudecode #claude #llm #promptengineering

19 min read

Murooka

Nov 8

Benign Injection: How Far a Single Line Like "Known Issue" or "Handled Elsewhere" Bends an AI's Judgment

#claudecode #agents #llm #security

24 min read

Murooka

Nov 8

Claude Said "Done as Instructed" but Wasn't Following Me. It Was Following Old Specs It Had Saved in Its Own Memory and Rules

#claudecode #claude #llm #agents

17 min read

Murooka

Nov 6

Is AI Review a Waste? I Counted My Real Work Logs, and About 60% of Findings Were Used

#claude #ai #codereview #productivity

20 min read

DEV Community

Murooka

Badges

Writing Debut

Claude Code Forgets My Instructions as Compactions Pile Up. One Added Prompt Stopped It: 0 Lost After 3 Compactions

Want to connect with Murooka?

Neither Forbidding nor Asking Reduced the Unwanted Behavior. Defining Where It's Allowed Did. Tested on Claude Code

Benign Injection: How Far a Single Line Like "Known Issue" or "Handled Elsewhere" Bends an AI's Judgment

Claude Said "Done as Instructed" but Wasn't Following Me. It Was Following Old Specs It Had Saved in Its Own Memory and Rules

Is AI Review a Waste? I Counted My Real Work Logs, and About 60% of Findings Were Used