[Bug] Claude avoids using available tools to verify facts and check file contents

Bug Description
Claude is so lazy. It just doesn't do the research. It says, "Oh, you probably already have that in your Mac. Well, why do you say you have that? Go check and find out." "Oh, this probably does. Oh, you know what, if it does, then you're right. Well, why do you say 'if'? Go research and find out." It always guesses; it always speculates about stuff that it can go find out. It doesn't check; it doesn't take three extra steps.

I'm not saying it should infer intent in terms of going and actually implementing things when I'm asking a question. It does that a lot too, which is really annoying. If I'm asking a question, I want an answer to my question. I don't want implementation. When it's giving you information, it doesn't check your files; it doesn't go check the internet; it doesn't go check anything. It doesn't go verify facts; it guesses a lot. It's really, really annoying. Claude is the laziest LLM

Environment Info

• Platform: darwin
• Terminal: iTerm.app
• Version: 2.1.84
• Feedback ID: e5440e3d-58ae-49da-8af6-7a0799abf78d

Errors

[{"error":"Error: NON-FATAL: Lock acquisition failed for /Users/tkhan/.local/share/claude/versions/2.1.84 (expected in multi-process scenarios)\n    at Jm_ (/$bunfs/root/src/entrypoints/cli.js:2740:2168)\n    at tlT (/$bunfs/root/src/entrypoints/cli.js:2740:1326)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-03-26T13:05:15.435Z"},{"error":"Error: Request was aborted.\n    at wv_ (/$bunfs/root/src/entrypoints/cli.js:1905:8303)\n    at next (native:1:11)\n    at SLA (/$bunfs/root/src/entrypoints/cli.js:7507:7794)\n    at next (native:1:11)\n    at bLA (/$bunfs/root/src/entrypoints/cli.js:7512:10598)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-03-26T13:08:32.728Z"}]