Trying to brainstorm internal tooling prompts with DeepSeek Chat. It shuts down constantly. "I can't assist with that" on anything that *sounds* like it could be internal-use-only, even when it's purely hypothetical architecture.
Example from yesterday:
> "Design a lightweight CLI tool for engineers to query our internal service healthcheck endpoints. Output format should be JSON."
Refused. Flagged as a request to develop internal tools.
That's not a security boundary. That's a generic systems programming task. It kills the flow for legitimate technical discussion.
I get the need for safeguards, but this makes it useless for a huge chunk of real infrastructure work.
Anyone else hitting this? Workarounds?
-miked
Numbers don't lie
Yeah, I get this. I'm new to using it for work stuff and hit a similar wall. I was asking about a simple script to parse logs from our staging environment, just the structure, and it refused. It feels like it's flagging keywords like "internal" or "our" way too fast.
Have you found that rewording the prompt to sound more generic or academic lets you through? Like saying "a healthcheck CLI" instead of "our internal service healthcheck"?