[BUG] Cowork VM ENOSPC persists across new sessions — Bash tool completely unusable

Bug Description

Cowork VM hits ENOSPC (no space left on device) and the error persists across new sessions. Two consecutive sessions (Feb 26 and Feb 27, 2026) have been completely bricked — no Bash commands execute at all.

Error: ENOSPC: no space left on device, mkdir '/sessions/[session-id]/tmp/claude-XXXX'

Environment

• Platform: macOS (Mac mini, Apple Silicon)
• Plan: Pro
• Claude Desktop: Latest version

Key Details

• Issue persists across new sessions — not a one-off
• /sessions/ workspace fills up and never gets cleaned between sessions
• No user-created files causing it — internal temp/cache bloat
• Affects ALL Bash tool usage (pip, git, python, npm — everything fails immediately)
• Two full sessions completely unusable

What I've Tried

• Starting new sessions (problem persists)
• Local cache clearing (~/Library/Application Support/Claude/Cache, vm_bundles, sessions)
• Force-killing Claude processes

Expected Behavior

New sessions should start with a clean /tmp directory, or there should be an auto-cleanup process between sessions.

Actual Behavior

The VM disk is full from prior session temp files. Since the Bash tool itself needs to write to /tmp to execute, no cleanup is possible from within the session — it's a catch-22.

Impact

Running 3 businesses through Cowork. Workaround (pushing code via GitHub API + Chrome browser tool) cuts throughput ~80%. Two bricked sessions = significant lost productivity.