Claude Code Says ENOSPC Or 0MB Free, But Your Mac Has Space

If Claude Code loses command output under a task capture path while df -h still shows plenty of free space, do not start by deleting Claude state. First prove whether this is real disk pressure, inode exhaustion, or a capture-layer failure.

Free first-pass triage

Turn one Claude Code ENOSPC incident into a safe first check.

Send a short public-safe summary or read-only scan first. If the incident needs a same-day cleanup order, the $29 review remains available after the risk boundary is clear.

task_dir="/tmp/claude-503"; df -h "$task_dir" "$HOME"; df -i "$task_dir" "$HOME"; du -sh "$task_dir" 2>/dev/null

Ask for safe next step Request $29 review Request $99 team policy Read-only evidence AI CLI runbook Open GitHub invoice

Before deleting Claude state

Get the safest next Claude ENOSPC check by email.

Email is enough for the first pass. Add the exact Claude warning only if you have it handy; we reply with the next read-only check before you delete any Claude state.

Evidence Before Cleanup

The key split is whether normal writes still work outside Claude's task output capture path. If $HOME writes succeed while the task directory reports 0MB free, broad cache deletion may not address the real failure mode.

task_dir="/tmp/claude-503"
df -h "$task_dir" "$HOME" 2>/dev/null
df -i "$task_dir" "$HOME" 2>/dev/null
du -sh "$task_dir" 2>/dev/null
find "$task_dir" -maxdepth 3 -type f -name "*.output" -print 2>/dev/null | head -40
printf "capture-test\n" > "$HOME/claude-output-workaround.txt"
ls -lh "$HOME/claude-output-workaround.txt"

Do Not Delete These First

~/Library/Application Support/Claude if you rely on local app state or VM bundles.
~/.claude if it contains sessions, plugins, auth, or active project state.
Project worktrees, transcripts, credentials, local databases, or mounted folders.

Copy-ready GitHub reply

Use this when Claude reports 0MB free but macOS has space.

This keeps the thread focused on proof instead of risky reset advice.

This looks like a task-output capture ENOSPC, not necessarily real disk exhaustion. Before deleting Claude state, I would collect the exact task-dir evidence and prove writes outside the harness still work:

task_dir="/tmp/claude-503"
df -h "$task_dir" "$HOME" 2>/dev/null
df -i "$task_dir" "$HOME" 2>/dev/null
du -sh "$task_dir" 2>/dev/null
find "$task_dir" -maxdepth 3 -type f -name "*.output" -print 2>/dev/null | head -40
printf "capture-test\n" > "$HOME/claude-output-workaround.txt"
ls -lh "$HOME/claude-output-workaround.txt"

If $HOME writes succeed while Claude says the task capture path has 0MB free, keep redirecting important command output outside the task dir and avoid broad deletion of Claude app support, ~/.claude, VM bundles, or active session state.

Request $29 review GitHub $29 invoice

Paid Review Scope

The $29 review is intentionally narrow: one Claude Code ENOSPC incident, one public-safe summary or read-only scan, and one safe/review/do-not-touch cleanup order. If the same incident affects a team, the $99 pilot turns it into a reusable policy.