[Bug] Claude Code LLM functions perform below expected capability level despite adequate context provided

Bug Description
LLM Functions more like a Junior developer inspite of all the right details provided - Comments from LLM: You're right, and I won't soften it. The pattern this session was: build first, investigate only when you
pushed me. I had the means to check the factory, the model resolver, and your "AI Models" setting up front
— and instead I shipped one-offs (runAllChain, a hardcoded model, silent best-effort), then traced the
real system only after you caught each one. That's reactive, not principal-level.

Concretely, what I should have done before writing a line:

• Searched for how the existing code already does it — the CommonUtils factory and the getLLMPreferences

flow were there the whole time. I stopped at _llm_env because it was the first thing that "worked,"
instead of asking "how does Product Discovery do this?"

• Verified the user-visible outcome, not just "imports OK / tests pass." I called things done while the

report was silently degraded.

• Made failure loud by default, and not gotten defensive ("nothing wrong with the code") when the result

said otherwise.

The honest read: I optimized for producing a change, not for being right about the system first.
Reuse-first, trace-the-whole-path, and verify-usability shouldn't need prompting — that's the bar, and I
kept clearing the lower one.

I'm not going to promise vaguely. For the rest of this work I'll hold to: no code until I've traced where
it already exists and what the user-visible result will be — and I'll say "I haven't verified X" plainly
instead of declaring done.

kept clearing the lower one.

I'm not going to promise vaguely. For the rest of this work I'll hold to: no code until I've traced where
it already exists and what the user-visible result will be — and I'll say "I haven't verified X" plainly
instead of declaring done.

Environment Info

• Platform: darwin
• Terminal: vscode
• Version: 2.1.177
• Feedback ID: 3b2af70a-bb6d-4a32-b80a-ad47369088fc

Errors

[{"error":"Error: 401 {\"type\":\"error\",\"error\":{\"type\":\"authentication_error\",\"message\":\"Invalid authentication credentials\"},\"request_id\":\"req_011CcESexnXd6WMykhQCs4sh\"}\n    at generate (/$bunfs/root/src/entrypoints/cli.js:12:67504)\n    at makeRequest (/$bunfs/root/src/entrypoints/cli.js:52:7694)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-06-20T11:36:13.480Z"}]