Version: 0.20.15
OS: Windows 11
Command: headroom learn --all --apply

Steps to reproduce

1. Have a Claude Code project in a directory whose path contains spaces
   (e.g. C:\Users\<name>\Desktop\Claude Code Projects)
2. Run headroom learn --all — dry run completes fine
3. Run headroom learn --all --apply — crashes

Error

FileNotFoundError: [WinError 161] The specified path is invalid: '\\\\\\Users\\AegeanPulse\\Desktop\\Claude'

Root cause

In headroom/learn/plugins/claude.py, the _decode_project_path() function (and the fallback in discover_projects()) reconstructs a Windows path from Claude Code's encoded project directory name by splitting on - and rejoining with \\:

drive = parts[0].upper()
win_path = Path(f"{drive}:\\" + "\\".join(parts[1:]))

Two issues:

1. Spaces in directory names are encoded as - (same as path separators), so Claude-Code-Projects is decoded as three separate path components instead of one directory named "Claude Code Projects".
2. The drive letter f-string (f"{drive}:\\") already ends with a backslash, making the path boundary \\\ (interpreted as UNC) rather than \\.

Expected behaviour

headroom learn --apply writes recommendations to the correct path:
C:\Users\<name>\Desktop\Claude Code Projects\CLAUDE.md

Workaround

Manually apply the recommendations printed by the dry run.