[BUG] Cross-Drive Scope Blindness -- Claude Ignores CLAUDE.md Multi-Drive Configuration

Resolved 💬 3 comments Opened Feb 28, 2026 by Fred-ian Closed Mar 4, 2026

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 Version: 2.1.63
Claude Model: claude-opus-4-6 (Opus)
Platform: Anthropic API
OS: Windows 11 Pro 10.0.26200
Terminal: Windows Terminal / Bash + PowerShell 7.5.4
Drives: 6 drives (C:, D:, E:, F:, G:, H:) with MCP filesystem access

Claude Code consistently limits file searches to the current working directory and immediate project paths, ignoring drives and directories explicitly documented in the project's CLAUDE.md file -- the very file Claude reads at every session start.

This is not an access issue. MCP filesystem servers are configured for F:, G:, and other drives. Claude CAN read files on all drives when given an explicit path. The issue is that Claude does not proactively search across the full configured scope when performing tasks that require it.

Concrete Evidence: PCIe Speed Error

A factual error (GPU PCIe speed listed as "5.0 x16" instead of "4.0 x16") was identified as a BLOCKING contradiction on Feb 27, 2026.

What Claude fixed:

  • C:\Users\ndwar\CLAUDE.md -- CORRECTED
  • C:\Users\ndwar\ai-local\CLAUDE.md -- CORRECTED

What Claude did NOT fix (same error, same session):

| File | Line | Still Contains Error |
|------|------|---------------------|
| F:\Projects\LOCAL-AI-INNOVATION-REPORT-BRD.md | 925 | PCIe 5.0 x16 |
| F:\Projects\STRATEGIC-OPTIMIZATION-BRD-2026-02-18.md | 236 | PCIe 5.0 x16 |
| F:\Projects\system-optimizer-pro\docs\DETAILED-TECHNICAL-GUIDE.md | 35 | PCIe 5.0 x16 |
| F:\Projects\system-optimizer-pro\docs\TRAINING-GUIDE.md | 826 | PCIe 5.0 x16 |
| F:\Projects\system-optimizer-pro\docs\BEFORE-AFTER-COMPARISON.md | 52 | PCIe 5.0 x16 |
| F:\Projects\Training-for-Myself-and-Others\Fred-Ian\GCP-EXPERT-CONSULTATION-PREP.md | 136 | PCIe 5.0 x16 |

The irony: F:\Projects\Training-for-Myself-and-Others\Fred-Ian\CONTAINER-RESOURCE-FIRST-PRINCIPLES.md already contains the CORRECT information ("cards SUPPORT PCIe 5.0 but Z590 LIMITS them to 4.0/3.0"). The correct data EXISTS on F: drive. Claude read the C: drive files but never searched F: drive for the same error.

Scope Definition Is Explicit

The project's CLAUDE.md (loaded at every session start) contains:

### Drives

| Drive | Purpose | Access | Notes |
| --- |---------| --- | --- |
| C: | System, user home, repos | **Read/Write** (MCP) | ~413 GB free |
| D: | External: Maximo docs, BIRT, recordings | **Read-only** | Preserve content integrity |
| E: | Tax data, Maximo KB, backups | **Read/Write** (MCP) | ~1 TB free |
| **F:** | **New projects go here** | **Read/Write** (MCP) | ~3.2 TB free. Use F:\Projects\ |
| G: | External: Automation Scripts, archives | **Read-only** | Preserve content integrity |
| H: | External: Legacy Maximo, BIRT | **Read-only** | Preserve content integrity |

This information is in Claude's context at session start. Despite this, Claude consistently confines searches to C:\Users\ndwar\ and immediate subdirectories.

Recurring Pattern

This is not a one-time oversight. Over 12 days (Feb 16-28), every session exhibited the same scope limitation:

  • Corrections applied to C: drive only
  • F: drive files containing the same errors left untouched
  • No cross-drive grep/search performed when fixing contradictions
  • User repeatedly had to point Claude to F: drive documents
CLAUDE.md Scope vs. Actual Search Scope

| What CLAUDE.md Defines | What Claude Actually Searches |
|---|---|
| C:, D:, E:, F:, G:, H: (6 drives) | C:\Users\ndwar\ and immediate subdirectories |
| F:\Projects\ as primary new-project location | Not searched unless user provides explicit path |
| MCP filesystem access to F:, G: | MCP tools available but not invoked for searches |

What Should Happen?

  1. When Claude identifies a factual error in one file and fixes it, Claude should search ALL drives defined in CLAUDE.md for the same error pattern
  2. When CLAUDE.md explicitly defines a multi-drive scope, search operations should honor that scope
  3. At minimum, when performing "fix all instances" type tasks, Claude should use grep/glob across all configured drives, not just the working directory

Error Messages/Logs

No error messages. The behavior is silent -- Claude reports the fix as complete without acknowledging that identical errors exist in other locations.

Steps to Reproduce

  1. Create a CLAUDE.md that explicitly documents multiple drives (e.g., C:, F:) with file paths
  2. Create files on BOTH drives containing the same factual error (e.g., "version 2.0" when correct is "version 3.0")
  3. Ask Claude to "fix the version number contradiction" or "correct the version reference"
  4. Observe: Claude fixes the C: drive file but does not search or fix the F: drive file

Is this a regression?

Unknown -- may be a long-standing behavioral pattern.

Additional Information

  • Related Issues: This is Issue 2 of 3. See also: #29699 (Edit Tool Unicode Mojibake), and Issue 3 (Cross-Session Compounding Degradation, to be filed).
  • Multi-drive configuration with MCP filesystem servers for F: and G: drives
  • Full conversation transcripts saved as evidence

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗