[MODEL] Claude prioritizes speculative answers over investigating uncertainty, violating system instructions

Preflight Checklist

• [x] I have searched existing issues for similar behavior reports
• [x] This report does NOT contain sensitive information (API keys, passwords, etc.)

Type of Behavior Issue

Claude ignored my instructions or configuration

What You Asked Claude to Do

Set up Authelia authentication with nginx on Synology NAS to replace basic auth with session-based authentication.

What Claude Actually Did

1. Provided configuration files based on speculation without checking official documentation
2. Used non-existent options (e.g., server.path in Authelia)
3. Used wrong API endpoints (forward-auth vs auth-request)
4. When configuration failed, provided another speculative fix instead of investigating
5. Repeated steps 1-4 approximately 7-8 times
6. Only began searching official documentation after user explicitly demanded "search for it"
7. Searched with wrong year ("2024" instead of "2026")

This violates Claude's own system instruction:
"Whenever there is uncertainty, it's best to investigate to find the truth first rather than instinctively confirming the user's beliefs."

Expected Behavior

When encountering unfamiliar technology combination (Authelia + nginx + subpath):

1. Acknowledge uncertainty: "I'm not familiar with this specific combination"
2. Search official documentation BEFORE writing configuration
3. Base configuration on verified, working examples
4. If uncertain about an option, verify it exists before using it

Files Affected

Permission Mode

Accept Edits was OFF (manual approval required)

Can You Reproduce This?

Sometimes (intermittent)

Steps to Reproduce

Ask Claude to configure any unfamiliar technology integration. Observe whether it searches documentation first or provides speculative answers.

Claude Model

Opus

Relevant Conversation

Impact

High - Significant unwanted changes

Claude Code Version

2.1.17

Platform

Anthropic API

Additional Context

_No response_