Claude Code kit · tested with Claude Code 2.1.310 on 2026-10-25
Not another markdown pack: hooks that run every session
Every session opens with your memory, git state and last handoff. Each prompt gets only the rules that match it. Claude can't stop on "you'll need to run…". API-credit workers stop at your daily dollar cap.
4 hooks · 6 commands · 6 skills · 17 tests you run yourself — installs with claude plugin install.
Lite: free (MIT). Pro: $49 once. Updates club: $9/month — a kit re-tested on each month's Claude Code release (Claude Code shipped 26 releases from 2026-09-25 to 2026-10-25).
1 · What your CLAUDE.md costs every session
2No-dodge check — paste Claude's last reply
The Stop hook reads Claude's final reply. If it hands work back to you, claims success it didn't check, leaves a placeholder or asks permission for the obvious next step, Claude is sent back once to do it — or to name the real blocker in one "Blocked:" line. English, 日本語, Deutsch, Español, Português.
3What you see in your terminal — real output
Captured by the kit's tests in a sample project, and from a real Claude Code 2.1.310 run with the plugin installed. Nothing retouched.
4What runs, and when
| Piece | When | What it does |
|---|---|---|
| Session-open summary | SessionStart | Date, git branch and uncommitted files, your MEMORY.md index, the last handoff note, rules, today's budget. After a long break: Claude Code's own estimate of the cache re-write cost. |
| Rule feeder | UserPromptSubmit | Only the rules.md sections and memories whose keywords appear in the prompt, right next to it. Not again for 10 prompts; again after /compact. |
| No-dodge | Stop | Sends Claude back once when its last reply dodges. Off while it is already continuing or background work runs; at most 3 times a session. |
| Budget guard | PreToolUse | Denies paid runs once today's cap is used; asks before a nested claude -p that could bill your subscription. |
| okit-fork | command | Background worker with claude --bare -p: API key required (bare mode never reads your subscription login), reserved in the budget first, capped in dollars and turns, real cost recorded. |
| okit-cache | command | Cache read share and requests per prompt from your own transcripts, before and after a date. |
| okit-memory · okit-rules · okit-budget · okit-doctor | commands | One fact per file with an index and a lint · import your CLAUDE.md into rules · daily/monthly caps · self-test of every hook. |
- claude plugin validate: passed (plugin and marketplace)
- claude plugin install from the marketplace: installed, enabled
- claude plugin details: ~569 tokens always-on for the 6 skills; hooks "no model context cost"
- This build session, measured with okit-cache: 98.7% of input tokens read from the prompt cache
5Lite, Pro, updates club
Lite — free
$0
- Session-open summary hook
- No-dodge Stop hook (5 languages)
- Memory template: MEMORY.md index + one fact per file
- Installer + tests (MIT)
Pro — the whole kit
$49 once
- Everything in Lite
- Rule feeder + rules template + CLAUDE.md import
- Budget guard + okit-budget ledger
- okit-fork API-credit workers (capped, worktrees)
- okit-cache, okit-memory, okit-doctor, 6 skills
Updates club
$9 /month
- Everything in Pro
- A new kit version each month, re-tested on that month's Claude Code release
- Claude Code shipped 26 releases from 2026-09-25 to 2026-10-25
- Cancel anytime: you keep every file you downloaded
Have a Pro or club key?
Enter it to download your CLAUDE.md converted into rules.md for the rule feeder. Only the key is sent to getreadystack.com to check it.
Questions
Can a fork bill my Claude subscription?
No. okit-fork always runs claude --bare, which never reads the subscription login, and refuses to start without an API key. It reserves the job's maximum in your daily budget first and records the real cost afterwards.
What do I need?
Claude Code and Python 3.8 or newer. macOS, Linux, or Windows with Git Bash. Plugin install, or --settings mode if you don't use plugins.
What leaves my machine?
Nothing. The hooks and commands are local scripts. The one network call is optional: checking a club key. This page also keeps your text in the browser.
Does it slow Claude down or eat tokens?
The hooks are local Python and finish in well under a second. Claude Code reports the six skills at about 569 tokens always-on. The rule feeder adds only the matching sections, once per 10 prompts. Measure your own sessions with okit-cache.
What happens when Claude Code changes?
Every kit version says which Claude Code version its tests ran on, and okit-doctor compares it with yours. The updates club gets a re-tested version every month.
Is this made by Anthropic?
No. ReadyStack is independent. Claude and Claude Code are Anthropic's products.
