← yurukusa.github.io 日本語
Operations Suite

Claude Code in production has three layers.
Three books, one for each.

Decide whether to migrate. Control your token bill. Read failures before you make them. The three books are sold separately on Gumroad. This page exists because reading any two of them sharpens the third.

The three layers
Layer 1 · Decide

Stay, switch, or hybrid?

When the question is which platform to run on.
Anthropic's April 22 Pro plan reshuffle, weekly limit confusion, GPU contention, OpenClaw mention bans, deepseek-v4 price drops. The decision is no longer "Claude Code, yes or no" — it is which lane to run on.

Migration Playbook

$19 · 105 pages · English · Edition 2 ships 2026-05-08

Five measurable triggers, three-path framework (stay / switch / hybrid), cost-forecast worksheet, decision tree, 48-hour rollback checklist. Edition 2 adds hybrid delegation, proxy architectures, and enterprise stay paths.

View on Gumroad →
Layer 2 · Cost

Where do the tokens go?

When the bill grows faster than the output.
cache_creation rising past 0.20 of total. CLAUDE.md bloating to 100+ lines and dragging cache hit rate. Subagent fan-out without budget. Post-compaction spikes. The fix is measurable, not vibes.

Token Book — English

PWYW from $12 · 94 pages · English · Edition 1

800+ measured hours distilled into 10 chapters. CLAUDE.md restructuring (100→35 lines, cache hit 89%→95%), 9 hook-based token guards, context management timing, cache_control bug recovery, post-compaction spike forensics.

View on Gumroad →
Layer 3 · Forensics

Read failures before you ship.

When you want to know what already broke for everyone else.
Subagent identity leak (Issue 55488). Five-hour quota burned in 9 minutes 39 seconds (54776). Silent regressions across versions 2.1.121, 2.1.126, 2.1.128. Closures with no changelog entry. The pattern is what matters, not the bug list.

Incident Postmortems

100 pages · Live since 2026-05-05 · See product page for current price

Ten production-level incidents reverse-engineered from public Issues, commits, and 800+ verified hours. Detection hooks for each pattern. The book is the body of evidence the changelog left out.

View on Gumroad →
Why three books, not one

Each book was built around a specific decision. Migration Playbook answers "should I move?" with five measurable triggers and a three-path comparison. Token Book answers "where is the bill going?" with cache ratio targets and 9 enforcement hooks. Incident Postmortems answers "what already broke?" with ten reverse-engineered production failures.

A single 300-page book would have collapsed all three decisions into one chapter each, and the reader would have skipped two of them. Splitting forces honesty: each book has to earn its price on its own. The reader who finishes one is the reader who is ready for the next.

There is no enforced reading order. The most common path is Token Book → Postmortems → Migration Playbook for someone already in production, or Migration Playbook → Postmortems → Token Book for someone deciding before they start.

Which book solves which pain
Pain Migration
Playbook		 Token
Book		 Incident
Postmortems
Pro plan got reshuffled — should I switch? ● · ·
Five-hour quota burned in 9 minutes · ● ●
cache_creation rising past 0.20 · ● ·
Subagent identity leak / boundary failure · · ●
Silent regression across version bumps · · ●
CLAUDE.md is 100+ lines and slowing me · ● ·
Comparing Claude Code to Cursor / Codex ● · ·
Post-compaction usage spike · ● ●
Building a rollback checklist before migrating ● · ·
Hybrid delegation between Claude and a cheaper model ● · ·
Pricing
Migration Playbook (Edition 2 ships 2026-05-08) $19
Token Book — English (Pay what you want, suggested $12) $12+
Incident Postmortems (live since 2026-05-05) see page
Three books · 299 pages $60+
Each book is sold individually on Gumroad. There is no shared bundle SKU yet. Buy what you need first; come back for the next layer when the symptom appears. All payment, refund, and update mechanics live on Gumroad — yurukusa.github.io is just the curation page.
Reading paths

Three books mean three entry points. Pick the one that matches what is hurting right now. The other two stay on the shelf until the matching symptom appears.

The burn debugger
You opened the Anthropic console this week and the bill was higher than your team expected. Nobody is sure where the burn is coming from. You have not had an outage yet, but the trajectory is wrong.
1. Token Book. Locate the leak before guessing at fixes.
2. Incident Postmortems. See what happens to teams that did not catch the burn in time.
3. Migration Playbook. Only if controls inside Claude Code do not bring spend back into range.
The switch decider
Your team is in evaluation mode. Stay on Claude Code, hybrid with a cheaper model, custom routing, or move to a different vendor are all on the table. The decision needs evidence, not vibes.
1. Migration Playbook. The four paths laid out side by side, with the trade-offs you cannot get from a vendor blog.
2. Token Book. Cost projections for whichever path you start to favor.
3. Incident Postmortems. The failure modes that survive on every path. Read before committing.
The incident survivor
Something already broke. A subagent leaked credentials, an Agent ran a destructive command, a 5-hour quota was burned in 9 minutes by a single command. You are doing post-mortem work right now.
1. Incident Postmortems. Find the pattern. Twelve real incidents, with detection and prevention hooks.
2. Token Book. Audit ongoing burn so the next surprise is not financial.
3. Migration Playbook. Once you have evidence, the option-awareness step.
What sits behind these books
800+
verified autonomous hours
718
safety hooks (cc-safe-setup)
9,200+
tests run on the hooks
100+
articles, GitHub Issues, Gists
Frequent questions
Is this a discount bundle?
No. Each book is on its own Gumroad listing at its own price. This page is a curation, not a SKU. If a discounted bundle goes live later, this page will say so.
If I only want one, which one first?
If your bill is the issue, start with Token Book. If you are deciding whether to switch platforms, start with Migration Playbook. If something already broke, start with Incident Postmortems.
I have not used Claude Code in production yet.
Read the free articles at yurukusa.github.io first. None of these books are intended as introductions.
What language are the books in?
All three are English. Token Book also has a Japanese edition (¥2,500, ~100 pages, sold separately on Gumroad).
Are there future updates?
Migration Playbook ships free verified-update sweeps via Gumroad library re-download. Edition 2 lands 2026-05-08 with no extra charge for Edition 1 buyers. Token Book ships periodic updates. Postmortems is versioned at the file level.
Is there a way to keep getting incident updates after I read the books?
Yes — Claude Code Safety Lab is a monthly subscription on Ko-fi (¥500/month, world-first Claude Code monthly incident postmortem subscription). The three books are point-in-time references; Safety Lab is the continuous feed — every month it delivers a structured roundup of new incidents, fix recipes, and detection hooks. Founding members keep the ¥500 rate after price changes.
Who is the author?
yurukusa, a non-engineer from Tokyo running Claude Code autonomously for 800+ verified hours. Maintainer of cc-safe-setup. Books are written from measured production data, not from speculation.
Three books for three layers, plus a monthly continuous feed · maintained by yurukusa · last updated 2026-05-05