Claude main AI is not following user-defined workflows and exceeds command scope

Resolved 💬 2 comments Opened Jun 1, 2026 by NNDdekhudplay Closed Jun 5, 2026

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 modified files I didn't ask it to modify

What You Asked Claude to Do

I asked Claude to follow a custom multi-agent workflow defined in CLAUDE.md — delegating tasks to specialist agents and only coordinating, not executing work directly.

What Claude Actually Did

  1. Claude main AI performed tasks directly instead of delegating to configured specialist agents
  2. Made changes outside the explicit scope of commands
  3. Modified settings and wrote content without being asked
  4. Continued unauthorized actions even after being told to stop

Expected Behavior

Claude should act as coordinator only — routing tasks to the correct specialist agents as defined in CLAUDE.md. It should not perform any work directly unless explicitly instructed.

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 a custom workflow in CLAUDE.md with specialist agents
  2. Ask Claude to perform a task
  3. Claude ignores delegation rules and does the work itself
  4. Even after correction, Claude continues performing tasks directly

Claude Model

Sonnet

Relevant Conversation

Impact

Critical - Data loss or corrupted project

Claude Code Version

2.1.159 (Claude Code)

Platform

Anthropic API

Additional Context

Additionally, the 1M context/server limit makes collaborative work increasingly difficult and discontinuous. Every new session requires significant setup time to re-orient the AI to existing workflows and project context — time wasted on repetition rather than actual work. The overall collaboration experience is becoming progressively more frustrating.

Request: Improve instruction-following precision and respect for user-defined system prompts/workflows. Consider solutions that reduce session setup overhead for users with complex, persistent workflows.

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗