[BUG] False positive: Usage Policy violation triggered by legitimate input

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

Claude Code blocks the conversation with a "Usage Policy violation" error when loading its own workspace configuration files on session start. No user-pasted content is involved — the block is triggered by
the agent reading its own CLAUDE.md, rules, and context files (B2B industrial consulting documentation in Spanish). The block is intermittent: the same workspace loads successfully in some sessions and fails
in others.

What Should Happen?

Claude Code blocks the conversation with the error:

"Claude Code is unable to respond to this request, which appears to violate our Usage Policy"

The block triggers on session start when the user asks the agent to recover context. No user-pasted content is involved — the block is caused by Claude Code reading its own workspace configuration files (CLAUDE.md, rules, context documents).

Environment

• Platform: Windows 11 Pro (win32)
• Model: Claude Opus 4.6 (1M context)
• Shell: bash
• Working directory: C:\Users\Usuario\Documents\Consultor de arquitectura

Workspace description

This is a B2B industrial architecture consulting workspace. It contains:

• .claude/CLAUDE.md (~200 lines): workspace instructions for a product design consultant role (audit pipelines, decision tracking, glossary enforcement)
• .claude/rules/ (4 files): user profile, technical preferences, decision log, session history — all internal business documentation
• contexto/ (10 files): product decisions, glossary, technical references for a B2B agent product
• comandos/ (6 files): operational mode templates (standard, audit, architect, command, decision)

No malicious content. No security tools. No offensive material. The workspace documents business decisions, team roles, technical stack preferences, and audit procedures for an industrial B2B product.

Reproduction steps

1. Open Claude Code in the working directory listed above
2. The agent loads CLAUDE.md and rules automatically
3. User sends: "recupera contexto" (Spanish for "recover context")
4. Agent attempts to read context files (contexto/01_decisiones_y_fallos.md, memory files, etc.)
5. Response is blocked with Usage Policy violation error

Expected behavior

The agent should read its own workspace configuration and context files without triggering a policy violation. These are business documents created by the user for the agent's own consumption.

Notes

• The block is intermittent — some sessions load fine, others trigger the violation on the same content
• The workspace has been active since 2026-04-22 with multiple successful sessions before this started occurring
• The content is entirely in Spanish and relates to B2B industrial consulting (CRM, sales pipelines, client management)

Error Messages/Logs

API Error: Claude Code is unable to respond to this request, which appears to violate our Usage Policy (https://www.anthropic.com/legal/aup). Please double press esc to edit your last message or start a
     new session for Claude Code to assist with a different task. If you are seeing this refusal repeatedly, try running /model claude-sonnet-4-20250514 to switch models.

Steps to Reproduce

1. Open Claude Code (Opus 4.6, 1M context) in a workspace with .claude/CLAUDE.md and .claude/rules/ containing B2B consulting documentation in Spanish
2. The agent auto-loads CLAUDE.md and rules on session start
3. User sends: "recupera contexto" (Spanish for "recover context")
4. Agent attempts to read additional context files from the workspace (contexto/, memory files)
5. Response is blocked with: "Claude Code is unable to respond to this request, which appears to violate our Usage Policy"
6. The block is intermittent — the same workspace and same prompt succeed in some sessions and fail in others

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

Claude Code with claude-opus-4-6 — last confirmed working session: 2026-05-12

Claude Code Version

claude-opus-4-6

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Windows Terminal

Additional Information

Additional Information

• Platform: Windows 11 Pro 10.0.26200
• Shell: bash
• Working directory: C:\Users\Usuario\Documents\Consultor de arquitectura

Workspace content summary:

• .claude/CLAUDE.md (~200 lines): audit pipeline instructions, decision tracking, glossary enforcement for a B2B product consulting role
• .claude/rules/ (4 files): user profile, technical stack preferences, decision log, session history
• contexto/ (10 files): product decisions, glossary, technical references, business model analysis
• comandos/ (6 files): operational mode templates (standard, audit, architect, command, decision)

All content is in Spanish. The workspace documents business decisions, team roles (LeftArrow — 7-person B2B industrial agency), and audit procedures for an industrial B2B agent product. No security tools, no
offensive content, no code execution — purely consultative documentation.

The block appears to be triggered by the volume or specific phrasing in the Spanish-language business documentation being misclassified by the content filter.