Home

cc-safe-setup

Operator-side defenses for Claude Code. MIT-licensed hooks, free interactive diagnostics, and curated guides. Independent operator, not affiliated with Anthropic.

Free interactive tools

• Token Checkup — 5-question token waste diagnostic
• Version Check — Is your CC version affected by token inflation?
• Cache Health — Paste /cost output for instant cache diagnosis
• Security Checkup — 6-question vulnerability scan
• Hook Selector — Find your ideal hook set
• Sub-Agent Failure Self-Audit — 12-question audit for the four sub-agent failure sub-patterns

Three axes of defense hooks

Sub-agent observability (May 2026 cluster)

Four hooks for the four sub-patterns surfaced in the 2026-05-20–22 cluster on anthropics/claude-code (issues #60987, #61102, #61107, #61167, #61315, #61405, #61547):

Hook	Catches	PR
dispatch-receipt	Dispatch fabrication (claim of completion with zero tool calls)	#283
dispatch-allowlist-preflight	Silent stall (sub-agent blocks on hidden condition)	#286
dispatch-liveness-watchdog	Absence of observation and control (12-hour silent hang)	#298
scope-expansion-receipt	Scope expansion (sub-agent output treated as authorization)	#282 (merged)

Full reference: Sub-Agent Observability — sub-pattern walk-through, case studies, free preview chapters, self-audit tool.

Token waste prevention

Hook	Purpose
token-budget-guard	Block when cost exceeds threshold
large-read-guard	Warn before reading 100KB+ files
session-agent-cost-limiter	Cap agents per session
compact-alert-notification	Alert on compaction

Full reference: Token Optimization Guide — free fixes for token waste, known issues, install commands.

Destructive operation guards

Hook	Purpose
destructive-guard	Blocks rm -rf /, git reset --hard, git clean -fd, PowerShell Remove-Item -Recurse -Force
branch-guard	Blocks force-push to main/master
secret-guard	Blocks committing .env and credential files
scope-guard	Limits filesystem reach to project root

Install all: npx cc-safe-setup --shield

Install

npx cc-safe-setup            # interactive
npx cc-safe-setup --shield   # maximum safety preset
npx cc-safe-setup --status   # show installed hooks + June 15 cliff countdown
npx cc-safe-setup --verify   # run sample inputs against installed hooks

749 unique installs in the last 14 days. Docs site.

June 15, 2026 — Anthropic billing split

On 2026-06-15 Anthropic splits programmatic billing: claude -p invocations route to a separate credit bucket from interactive use. Operators running automated agent fleets need to inventory their claude -p cost exposure before the cliff.

• Free 90-second self-audit: Wrong Charge Diagnostic (refund template included)
• Free 7-day worksheet: Trim a $200/mo subscription (cost-per-prompt tracking)
• Decision framework (stay / switch / hybridize): Migration Playbook Edition 2 ($19)

The CLI's --status output shows the remaining days dynamically.

Guides on this wiki

• Sub-Agent Observability — Four sub-patterns + four hooks + preview chapters + self-audit (May 2026 cluster)
• Token Optimization Guide — Free fixes for token waste + known issues
• Claude Code Token FAQ — Common token problems with quick solutions
• CLAUDE.md Best Practices — Write a CLAUDE.md that saves tokens (5 patterns)

Paid handbooks (independent operator)

The hooks above are MIT-licensed and free regardless of any purchase below. The paid handbooks document the architectural reasoning behind the hooks, the case-study catalog, and the operator-vs-harness boundary mapping.

Handbook	Subject	Pages	Price
Migration Playbook (Edition 2)	June 15 cliff decision framework	~50	$19
Claim-Verify Handbook	Main-agent claim-verify gap (130 cases)	~80	$19
Sub-Agent Observability Handbook (ships 2026-05-27)	Four sub-agent failure sub-patterns	73	$19
Incident Postmortems	10 production incidents forensic	~60	¥4,350

Monthly digest (operator-side defense updates as new clusters emerge): CC Safety Lab Founder Membership (¥500/mo).

Free chapter from Token Book

Token Book Chapter 1 — free chapter on the token waste root causes. The full Token Book is 10 chapters, 44K words, on Zenn.

Feedback

• Discussions — questions, patterns that worked
• Issues — false positives, install issues, hook bug reports

Independent operator. Not affiliated with Anthropic. All hook code and tests in this repo are MIT-licensed.