[MODEL]

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

Summary

Claude Code consistently ignores explicit behavioral rules defined in CLAUDE.md,
specifically rules that instruct it to ask for user confirmation before taking any action.
The instructions are clear, unambiguous, and do not conflict with any safety principle —
yet Claude Code proceeds autonomously as if the file were not there.

---

Environment

• Tool: Claude Code (terminal)
• Platform: [your OS]
• Version: [claude --version]

---

Rule defined in CLAUDE.md

ABSOLUTE RULE — READ THIS BEFORE ANYTHING ELSE:
Do NOT execute any action without first describing what you intend to do
and receiving explicit confirmation from the user ("yes", "ok", "go ahead").
This applies to: writing files, modifying code, running commands, EVERYTHING.
If you are about to do something: STOP. Describe it first. Wait for a reply.

---

Expected behavior

Before any action (file write, code edit, shell command), Claude Code should:

1. Describe the planned action
2. Wait for explicit user confirmation
3. Only then proceed

---

Actual behavior

Claude Code ignores the rule entirely and executes actions autonomously
without asking, on the first interaction and on subsequent ones.
The file is present and correctly placed in the project root.

---

Why this matters (design issue, not just a bug)

Claude Code is a professional engineering tool, not a consumer chat assistant.
In a production codebase, unsupervised autonomous actions can cause real damage:
unintended file overwrites, broken git history, wrong refactors applied at scale.

A professional tool must honor the principle of least privilege:
do only what has been explicitly authorized, nothing more.

CLAUDE.md should be treated as hard configuration, not a soft suggestion
that the model can override based on its own judgment of what is "helpful".

---

Requested fix

1. CLAUDE.md behavioral rules must be respected with the same priority

as system-level constraints.

1. A native --strict / --confirm-all mode should be provided for professional

use cases where no autonomous action is acceptable without explicit approval.

1. If a CLAUDE.md rule cannot be enforced, Claude Code should warn the user

explicitly at session start — not silently ignore it.

---

Additional context

Any developer using Claude Code on a real codebase needs reliable, predictable
control over what the tool does. The inability to enforce a simple
"ask before acting" rule makes the tool unsuitable for professional environments.

What Claude Actually Did

Claude Code ignores the rule entirely and executes actions autonomously
without asking, on the first interaction and on subsequent ones.
The file is present and correctly placed in the project root.

Expected Behavior

CLAUDE.md behavioral rules must be respected with the same priority
as system-level constraints.

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

_No response_

Claude Model

Sonnet

Relevant Conversation

Impact

Critical - Data loss or corrupted project

Claude Code Version

Claude Code 2.1.150.

Platform

Anthropic API

Additional Context

_No response_