Claude Code does not consistently follow user-defined conversation principles

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

I defined specific conversation principles (rules) that Claude must follow throughout the session. These included: always summarize instructions before acting, wait for approval, report only key points briefly, and follow user direction strictly.

What Claude Actually Did

1. Claude reverted to its default behavior patterns, ignoring the user-defined principles after a short time.
2. Claude made arbitrary modifications and additions without user instruction.
3. Claude asked unnecessary questions instead of simply following instructions.
4. The microphone (voice input) feature does not work in the PC version.

Expected Behavior

1. Claude should strictly follow user-defined principles throughout the entire session without reverting to default behavior.
2. Claude should only do exactly what the user instructs — no arbitrary additions or modifications.
3. Claude should execute precisely under user direction, not act on its own judgment.
4. The microphone (voice input) feature does not work in the PC/Windows version of Claude Code.

Files Affected

Permission Mode

Accept Edits was ON (auto-accepting changes)

Can You Reproduce This?

Yes, every time with the same prompt

Steps to Reproduce

1. Define conversation principles at the start of a session (e.g. always ask before acting, report briefly, do only what is instructed)
2. Continue the conversation for several exchanges
3. Claude gradually reverts to default behavior, ignoring the defined principles
4. The microphone (voice input) feature does not work in the PC/Windows version of Claude Code.

Claude Model

Sonnet

Relevant Conversation

User defined 16 specific principles at the start of the session.
Claude acknowledged them but gradually stopped following them as the conversation continued.
Claude added unrequested content, asked unnecessary questions, and made arbitrary modifications.

Impact

High - Significant unwanted changes

Claude Code Version

2.1.87 (Claude Code)

Platform

Anthropic API

Additional Context

This happens consistently throughout every session, not just occasionally.
The behavior reverts to default patterns after several conversation exchanges.
Even when principles are injected automatically via hooks on every user message, Claude still gradually ignores them.
Also: microphone (voice input) feature does not work in the PC/Windows version of Claude Code.