ml0x.com / Claude Code workflow pack
A Claude Code setup that plans first, cannot wreck your repo, and stops before it burns your budget
Five CLAUDE.md templates, five hooks and eight slash commands, as plain files you copy into .claude/. Markdown and standard-library Python, no account, no network calls. Read every line before you use it. That is why it ships as source.
$29 once. One zip, MIT licence, every project you own.
The pack is written and its hook tests pass (64 checks). Pay with Stripe, then you land on the ml0x download page with the zip.
Get the pack, $29What is in the zip
claude-md/ CLAUDE.global.md your rules for every project
CLAUDE.base.md any repo: commands, layout, traps, done
CLAUDE.python-ml.md PyTorch: shape rules, GPU-cost rules
CLAUDE.typescript-web.md
CLAUDE.subdir-example.md monorepo packages
hooks/ guard_bash.py protect_files.py cost_guard.py
format_on_edit.py notify.py
commands/ plan review test-fix commit pr
explain-error handoff secrets-check
settings/ settings.json (wires every hook)
tests/ test_hooks.py install.sh README.md
The five hooks
Claude Code runs a hook before or after each tool call. A PreToolUse hook that exits with code 2 blocks the call, and its message goes back to Claude as the reason, so the agent changes course instead of retrying.
| Hook | What it does |
|---|---|
guard_bash | Blocks rm -rf of /, ~ or ., force push to main, git reset --hard, git clean -f, curl | sh, sudo, DROP TABLE, terraform destroy, npm publish, cat .env and more. Allows the safe versions, like rm -rf ./build or a force push to a feature branch. |
protect_files | No edits to .env, keys, lockfiles, node_modules/, .git/, files outside the project, or the hooks and settings themselves. An agent should not be able to switch off its own guard. |
cost_guard | Per-session brake on tool calls (400), minutes (120), transcript size (8 MB) and the same command run 5 times in a row. Warns at 80 %, then tells Claude to write a summary and hand back. |
format_on_edit | Runs ruff or black, project prettier, gofmt, rustfmt or shfmt on each changed file. Never blocks. |
notify | Desktop notification when Claude needs a permission or finishes, so you stop watching the terminal. |
The eight slash commands
| Command | What it does |
|---|---|
/plan | Reads the code and writes a numbered plan with files, checks and risks. Edits nothing. Waits for "go". |
/review | Reviews the diff for bugs, security, API breaks and missing tests, each with file:line and an input that triggers it. |
/test-fix | Fixes failing tests one at a time, root cause first. Max 3 tries per test. Never skips or loosens an assertion. |
/commit | Commit message in your repo's style that says why. No AI attribution lines, never --no-verify. |
/pr | PR title and body: what, why, how to test, risk, follow-ups. |
/explain-error | Finds the first stack frame in your code and explains the cause before any edit. |
/handoff | Writes HANDOFF.md so you can /clear and continue in a fresh, cheaper context. |
/secrets-check | Scans added lines for keys, tokens and personal data, masked, and tells you when to rotate. |
Free sample: the /plan command, complete
This is one of the eight files, complete (long lines wrapped to fit here). Save it as .claude/commands/plan.md and type /plan in Claude Code.
--- description: Read the code and write a step-by-step plan before any edit. Makes no changes. argument-hint: <what you want built or fixed> allowed-tools: Read, Grep, Glob, Bash(git status:*), Bash(git log:*), Bash(git diff:*), Bash(ls:*) --- Task: $ARGUMENTS You are in planning mode. Do not edit, create or delete any file in this turn. 1. Restate the task in one sentence. If it is ambiguous, list the open questions and stop. 2. Find the code that the task touches. Read it. Name each file and the function or section, with line numbers. 3. Check how the project already solves similar problems (patterns, helpers, tests) and reuse them. 4. Write the plan as a numbered list. For each step give: the file, the change, and how you will verify it (a test, a command, or a manual check). 5. List the risks: what could break, which callers depend on the code you change, and any migration or data change. 6. Give a size estimate: number of files, rough lines changed, and whether the change should be split into more than one commit. End with: "Reply 'go' to implement, or edit the plan." Then wait.
The allowed-tools line is what makes it safe: the command can read and search, but it has no Edit or Write tool, so "plan only" is enforced, not requested.
Works in cloud sessions too
A Claude Code cloud session reads the .claude/settings.json committed in the repo, not your laptop's user settings. The pack installs at project level, so the same guards and commands run there. More on what carries over in the cloud sessions guide.
Pack or Pipeline
Workflow Pack, $29
- CLAUDE.md templates
- 5 hooks, 8 slash commands
- For: your own interactive Claude Code sessions
ml0x Pipeline, $99
- The 100x CLI, four-stage run with quality gates
- Dollar budget ceiling per run, parallel research runner
- For: unattended, multi-agent runs
Questions
Is this official Anthropic software?
No. It is an independent set of files built on Claude Code's documented hooks, settings and slash command features.
Does cost_guard know my real bill?
No, a hook cannot see billing. It limits tool calls, time, transcript size (a proxy for context, which drives per-turn token cost) and repeat loops. Use it with the spend limits in your Anthropic account, not instead of them.
What do I need?
Claude Code and Python 3 (for the hooks). macOS or Linux. The commands are Markdown and work anywhere Claude Code runs.
What happens after I pay?
Stripe sends you back to ml0x.com/welcome/, which checks the payment and shows the download for the zip. One payment of $29, no subscription.
Get the pack, $29 once
5 CLAUDE.md templates, 5 hooks and 8 slash commands as one zip. Download right after payment.
Get the pack, $29