Roll out Claude Code safely across your team

NOT ACCEPTED — a record of the org-level services that used to be offered here. The free material and the hooks continue.

Using Claude Code as an individual and using it across a team are not the same safety problem. In an organization, one developer's unprotected setup is the whole org's risk. An AI that runs fine on one machine can recursively delete files, stage .env into git, drop a production database, or trigger unexpected billing — and that happens no matter how careful the other nineteen people are. This page lays out how to make safety uniform, enforced and audited across the org, with the practical patterns to get there.

Sound familiar?

Training and a hardened environment don't stop the accidents

Most corporate Claude Code support is either training (teaching people how to use it) or environment hardening (no-training data settings, SSO, isolated environments). Both matter as an entry point. But the real accidents happen on the machine of someone who was trained, inside a hardened environment.

What this service adds is the layer after that: operational guardrails. Instead of relying on human attention, dangerous operations are stopped mechanically, before they run — recursive file deletion, production-database destruction, .env staging, force‑push, runaway sub-agents, unexpected billing. These are blocked with hooks, the same baseline is enforced across every repo in the org, and CI plus an audit trail keep confirming it still works.
Typical training / hardeningThis service (operational accident prevention)
FocusLearning to use it / securing the environmentStopping dangerous operations before they run
Depends onTrainee attention / following rulesMechanical enforcement (hooks, CI, audit)
Pricing shapeMainly per-seat training feesAudit + monthly review (not per-seat)
PositionThe on-rampThe downstream layer for the "accidents still happen in the field" problem that always remains

It composes with training rather than competing with it: the same diff-style enforcement that catches a weakened baseline in CI is what training alone can't give you.

The basis is real, documented incidents and on-machine verification of the hooks that stop them — for example an unauthorized transfer #46828 ($1,446), mass deletion of database tables #27063, and a secret leak that walked straight past CLAUDE.md instructions #2142. Not generalities — incidents you can trace by number, with the matching hook.

Services & indicative pricing

August 2026 notice: items 1–5 below are not being accepted

The org-level audit (¥150,000+), training, rollout, monthly retainer and spot consulting all require a person to keep doing the work. That capacity isn't something I can responsibly commit to, so none of them are currently accepted. If you looked at this page earlier and were considering it, I'm sorry. (Three written audits are a different shape and are still offered, because none of them needs a person to keep showing up: the CLAUDE.md Audit reads your instruction files and the Token Burn Audit reads your session logs and /cost output — $29 / ¥3,980 each, within 48 hours — while the Full-Surface Audit crosses CLAUDE.md, settings.json, your hooks, your session logs and your CI config against each other and reports only the contradictions that fall between the layers — $219 / ¥29,800, within 72 hours. All three are asynchronous, with no call and nothing run in your environment. See SERVICES.md, or read the full sample report the Full-Surface Audit produced against my own setup.)

The prices are left visible on purpose rather than quietly deleted — what was published shouldn't silently disappear. What can't be delivered is stated as not delivered.

Besides the three audits above, two things need no inquiry and no order at all: the Team Safety Rollout Pack (¥3,000), a one-off download, and the free material further down this page. The hooks themselves stay free (MIT).

The figures below are the estimates as they were published.

1. Safety audit (one-off)not accepted

Review the org's settings.json, CLAUDE.md and hook setup, and surface the gaps that lead to accidents (permission holes, secret-leak paths, deletion and billing risk). You get the remediation config and a distributable safety baseline. Delivered as a written report — no live session required.

from ~$1,000 (one-off) Deliverable: vulnerability report + recommended hooks + distributable template

See a sample vulnerability report → (fictional company, for illustration)

2. Team training (online / recorded)not accepted

For tech leads and developers: how to operate Claude Code safely, taught through real incidents and how to prevent them, mapped onto your own workflow. Available recorded so it scales across time zones.

from ~$350 / person (varies by size & format) Recorded option · 7-module curriculum

See the curriculum (7 modules) →

3. Rollout & setup (org baseline design + distribution)not accepted

Design the shared safety policy, the mechanism to distribute it to every repo and developer, the CI safety gate, and enforcement of the required baseline. Eliminates the drift and "forgot to add it" failures of manual distribution.

from ~$2,000 (one-off project) Deliverable: shared policy + CI workflow + rollout runbook

4. Monthly review & supportnot accepted

The core service. Ongoing hook updates, responses to newly reported incident classes, config review, and operational Q&A. Claude Code ships fast and the safety assumptions shift; training and initial setup are one-time, but the risk stays for as long as you keep using it. Priced per org's operation, not per seat.

from ~$700 / month (varies by scale) Continuous safety updates + review + audit

5. Spot consulting (design review / incident post-mortem)not accepted

One-off: AI-coding adoption strategy, safety-design review, or root-cause analysis and prevention after an incident has already happened.

from ~$2,000 / day Strategy / design review / post-mortem

6. cc-safe-setup for Teams (open-core · gauging interest)

The core hooks stay free and MIT, always. On top of that, a per-developer monthly tier that distributes and enforces a shared policy across the org, runs it in CI, and keeps an audit trail. I'm confirming demand before building it — if it'd help your org, a 👍 or a line is enough.

~$12 / developer / month (gauging interest) Deliverable: monthly compliance report + incident-rule feed + bypass audit

See a sample monthly compliance report → (fictional company, for illustration)

Demand thread (GitHub Discussion #632) →

Free, start right now

Before any conversation, check where you stand with the free tools — all usable in-house as-is.

ToolWhat it does
Team Governance ScorecardScore 8 org-level controls in 30 seconds; get a shareable card and a clear "what to fix next" list
Safety Scorecard (per developer)Check the individual layer: 8 documented incidents, X/8 coverage you can screenshot
cc-safe-setup (free · MIT)One command installs the core safety hooks. The individual tier stays free.

What is available

As stated above, the org-level audit (¥150,000+), training, rollout, monthly retainer and spot consulting are not being accepted, and the inquiry form has been closed. Leaving it open would mean collecting requests I can't answer. (Three written audits are a different shape and are still offered, because none of them needs a person to keep showing up: the CLAUDE.md Audit reads your instruction files and the Token Burn Audit reads your session logs and /cost output — $29 / ¥3,980 each, within 48 hours — while the Full-Surface Audit crosses CLAUDE.md, settings.json, your hooks, your session logs and your CI config against each other and reports only the contradictions that fall between the layers — $219 / ¥29,800, within 72 hours. All three are asynchronous, with no call and nothing run in your environment. See SERVICES.md, or read the full sample report the Full-Surface Audit produced against my own setup.)

Besides the three written audits above, these two need no inquiry and no order:

Team Safety Rollout Pack ¥3,000 (BOOTH, one-off) First, score your team (free)

The hooks themselves (cc-safe-setup, MIT) stay free. Bug reports and questions are still welcome through the normal issues and discussions.

About

Author of cc-safe-setup (914 example hooks · MIT) 2,534 clones / 628 de-duplicated in the last 14 days (measured; includes CI and mirrors — not a people count) 96 unique page viewers in the last 14 days (measured) GitHub 4★ 170k+ cumulative Qiita views on Claude Code safety & operations Designed from real, documented incidents (GitHub issues)

Figures were re-measured on 2026-08-09. Two of them were corrected in the process. "GitHub 42★" was wrong — the actual count is 4★, so the page had been overstating it by 10×. And "~30k npm installs", which used to appear here, has been removed: that number includes bot and mirror traffic, as I had already concluded elsewhere, so it can't stand as evidence of real users. The clone figure is now written so it's clear it isn't a count of people. No exaggerated promises.