Cluster Exposure Diagnostic

Which of the 16 tracked plus 7 candidate Claude Code failure clusters are most likely hitting you right now?

What this is. Over the seven-week window from mid-April through 2026-06-01, 16 tracked plus 7 candidate structural failure clusters surfaced in the public anthropics/claude-code issue tracker — 12,000+ reactions across 480+ open issues. None has an official upstream fix yet. Operator-side defenses ship continuously: as of 2026-06-01, 13 of the 16 tracked clusters and 5 of the 7 candidate clusters have at least one shipped defense hook in cc-safe-setup, with new hooks landing within 24-48 hours of each cluster's emergence.

What this diagnostic does. Seven multi-select questions about your Claude Code setup map to specific clusters. The result lists the clusters most likely affecting your workflow, with concrete hook installs and reading material per cluster.

Three minutes. Local-only. No tracking.

1. Which Claude Code surfaces do you use regularly?

Select all that apply.

CLI (terminal claude command)
Desktop app (Claude.app)
Cowork (Desktop sandbox / remote control)
Web (claude.ai)

2. Do you also use other coding agents on the same repositories?

Codex, Cursor, Amp, Aider, GitHub Copilot, etc. Select all that apply.

OpenAI Codex (CLI or IDE)
Cursor
Amp / Aider / Copilot CLI / others
Claude Code only

3. Do you manage multiple Claude accounts on the same machine?

For example, a personal account and a work account. Or two Pro Max subscriptions across teams.

Yes — I switch between accounts
No — single account only

4. Have you hit any of these in the last 30 days?

Select all that apply.

Pro Max quota burned faster than usage suggests
"malformed and could not be parsed" error on tool calls
An ordinary prompt blocked by usage policy (AUP)
Auto-compact silently failed to fire near token limit
None of these

5. Do you actively use sub-agents (Task tool spawns)?

For research, parallel work, or code review delegation.

Heavy use — multiple sub-agents per session
Occasional use
Rare or never

6. Has a permission rule in settings.json ever surprised you?

For example, an allow rule that didn't match, a deny that was bypassed, or a session-approval persisting unexpectedly.

Yes — at least once
No — rules behave as expected
I don't use custom permission rules

7. Do you use Claude Skills?

The ~/.claude/skills/ or project .claude/skills/ directories.

Yes
No
Not sure what these are

Your cluster exposure

Each card below is a cluster you are likely exposed to based on your answers, ordered by probability. Cards highlighted red are higher-probability for your setup; yellow are moderate; green are lower-priority background risks.

Fix it now, then go deeper

Install the deterministic operator-side defenses for these clusters in one step — they ship in the free, MIT-licensed cc-safe-setup. If you scored high on the sub-agent clusters (fabricated "completed" reports, silent stalls, scope expansion), the four sub-agent failure patterns are covered end-to-end — each with its install-ready hook — in a dedicated book (Japanese), chapter 1 free.

Install the free hooks: npx cc-safe-setup → · 副の作業者の沈黙の失敗 (¥1,500) — chapter 1 free →