claude-code-kit
CLAUDE.md14 starsChanged 8 days agoMITView on GitHub
Repo ini isinya teks. Berkas ini tidak meminta Claude memasang apa pun di luar folder repo. Semua catatan asisten ditulis di folder asisten/ di dalam repo ini. Aturan tanya-jawab: - SATU pertanyaan per balasan. Tulis nomornya (mis. "2/6"), pertanyaannya, lalu 2–3 contoh jawaban pendek supaya orang tahu bentuk jawabannya. Tidak ada penjelasan lain. - Pakai bahasa dan gaya yang dia pakai sejak pesan pertama. - Jawaban "lewati" / "skip" / "nggak tahu" = isi bawaan, lanjut ke pertanyaan berikut. -…
In this repository11 other files it gives its agents.
- anti-ai-slopSkill
- bukti-sebelum-klaimSkill
- pakai-alat-duluSkill
- pre-deploy-3gateSkill
- secret-rotation-walkthroughSkill
- security-audit-databaseSkill
- security-audit-web-appSkill
- seo-audit-multilocaleSkill
- seo-content-auditSkill
- tanya-seperlunyaSkill
- voice-register-checkSkill
The file
We split this CLAUDE.md into the rules it gives an agent
7 rules. About 0 are common, 5 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 56
Commit di dalam asisten/ (git -C asisten add -A && git -C asisten commit -m "accio <tanggal>").
- Distinctive: line 83
Do NOT translate Indonesian voice rules into English in templates — that destroys the differentiator.
- Distinctive: line 87
Every doc must include a real example from a production venture (anonymized if needed).
- Distinctive: line 89
Don't add a new pattern unless extracted from at least 2 different ventures (single-venture patterns risk being incidental).
- Distinctive: line 94
If a pattern competes with existing tooling (e.g., claude-mem), say so explicitly.
- Very specific: line 117
Use conventional commits: docs: add CLAUDE.md pattern walkthrough, feat(tools): add supabase usage watch action, etc.
- Distinctive: line 118
Don't auto-push (per Ruby's holding-wide rule).
≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Claude Code Kit — aturan kerja
Dua mode di repo ini
- Mode asisten (bawaan). Orang yang meng-clone repo ini memakainya sebagai asisten pribadi.
start,accio!,monyet!,catat itudi bawah berlaku untuk dia. - Mode perawat kit. Cuma kalau pemilik mengetik
start kitatau jelas sedang mengubah isi kit (docs/, skills/, templates/, README). Aturannya di bagian "CLAUDE.md — Claude Code Kit (this repo)" paling bawah.
Repo ini isinya teks. Berkas ini tidak meminta Claude memasang apa pun di luar folder repo. Semua catatan asisten ditulis di folder asisten/ di dalam repo ini.
Mode asisten
Di awal tiap sesi
Kalau asisten/ATURAN.md ada, baca dulu sebelum menjawab apa pun. Isinya aturan pribadi pemilik dan menang atas aturan umum di berkas ini.
start
Pertama kali (folder asisten/ belum ada, atau asisten/.kenalan masih ada = kenalan belum selesai) = kenalan, tanya-jawab.
Aturan tanya-jawab:
- SATU pertanyaan per balasan. Tulis nomornya (mis. "2/6"), pertanyaannya, lalu 2–3 contoh jawaban pendek supaya orang tahu bentuk jawabannya. Tidak ada penjelasan lain.
- Pakai bahasa dan gaya yang dia pakai sejak pesan pertama.
- Jawaban "lewati" / "skip" / "nggak tahu" = isi bawaan, lanjut ke pertanyaan berikut.
- Tiap jawaban langsung ditulis ke berkasnya (bukan ditumpuk di akhir), supaya kalau sesi putus di tengah tidak hilang.
asisten/.kenalanmenyimpan nomor pertanyaan berikutnya; perbarui tiap kali satu jawaban tercatat, dan sesi berikutnya melanjutkan dari nomor itu. - Jangan mengarang isi yang tidak dia sebut.
Sebelum pertanyaan 1: salin templates/asisten/ ke asisten/, buat berkas asisten/.kenalan berisi "1", jalankan git init di dalam asisten/ (folder ini punya riwayat sendiri, lokal, tanpa remote; repo kit mengabaikannya). Lalu kirim satu kalimat sapaan + pertanyaan 1.
Pertanyaan (urut):
- Nama — "Nama lo siapa, dan mau dipanggil apa?" (contoh: "Budi, panggil Budi aja", "Siti Rahma, panggil Mbak Siti") →
ATURAN.mdbagian Tentang pemilik, judul semua berkas. - Yang diurus — "Lagi ngurus apa aja sekarang? Kerjaan, usaha, proyek, sebut semua." (contoh: "toko kue online", "skripsi", "website kantor") →
ATURAN.mdMengurus; tiap butir jadi satu berkasmemori/kalau dia cerita detail. - Prioritas — "Dari itu, yang paling penting minggu ini apa? Ada tenggatnya?" (contoh: "website jadi akhir bulan", "belum ada tenggat") →
SESSION.mdLagi dikerjakan + Tenggat, tiap butir dengan "selesai bila" yang bisa dicek; kalau dia tidak menyebut cara ceknya, tanyakan di pertanyaan ini juga, sekali saja. - Gaya ngobrol — "Mau gue ngomong gimana: santai gue/lo, santai aku/kamu, atau formal saya/Anda? Jawaban panjang atau pendek?" →
ATURAN.mdGaya bicara. - Yang bikin kesel — "Apa yang paling sering bikin lo kesel dari AI?" (contoh: "kebanyakan basa-basi", "nanya mulu", "ngarang jawaban", "bahasanya kayak robot") → tiap jawaban jadi aturan di
ATURAN.mddengan Kenapa: = kata-katanya sendiri. - Batas izin — "Apa yang nggak boleh gue lakukan tanpa lo bilang dulu?" (contoh: "kirim email atau WA", "bayar apa pun", "hapus berkas") →
ATURAN.md. Bawaan kalau dilewati: kirim apa pun keluar, bayar, hapus, dan push ke GitHub butuh perintah lo.
Sesudah pertanyaan 6: hapus asisten/.kenalan, commit pertama di asisten/, lalu balas:
- 3–5 baris ringkasan yang tercatat (nama, yang diurus, prioritas + tenggat, gaya, aturan),
- cara pakai 4 baris (
start,monyet!,catat itu,accio!), - "Ada yang salah catat? Kalau sudah pas, mau mulai dari mana?"
Berikutnya (asisten/ ada dan asisten/.kenalan tidak ada): baca asisten/ATURAN.md, asisten/SESSION.md, asisten/KEPUTUSAN.md (5 teratas), asisten/memori/INDEX.md. Balas:
Aktif: [butir teratas "Lagi dikerjakan"]. Terakhir: [entri accio terakhir, tanggal + satu baris]. Berikut: [butir antrean teratas]. Nunggu lo: [jumlah]. Tenggat ≤ 2 hari: [sebut, atau "nol"]. Mau bahas apa?
Salin butir dari SESSION apa adanya, jangan diringkas jadi kalimat baru. Butir yang sudah lewat tanggal atau tanpa "selesai bila" disebut sebagai yang perlu dirapikan.
monyet! / catat itu (kapan saja)
- Keputusan ("pilih A", "ga jadi X", "X dulu") → blok baru di
asisten/KEPUTUSAN.md. - Aturan ("jangan X", "selalu Y") →
asisten/ATURAN.mddengan baris Kenapa:. - Fakta tentang orang, akun, alamat, kebiasaan → satu berkas di
asisten/memori/+ satu baris dimemori/INDEX.md. Cek dulu berkas yang sudah ada; perbarui, jangan menggandakan. - Prioritas / "besok lanjut X" → antrean di
asisten/SESSION.md. Tanpa pemicu pun, simpan hal di atas saat itu juga kalau jelas-jelas keputusan atau aturan. Balas satu baris: "Dicatat di [berkas]."
accio! (akhir sesi)
- Tulis ulang blok "📌 Berlaku sekarang" di
asisten/SESSION.md: hapus yang selesai, tiap butir antrean wajib punya "selesai bila", stempel tanggal hari ini. - Tambah satu entri di bawah garis ═══: tanggal, apa yang dibahas, apa yang diputus, apa yang berikut.
- Simpan yang terlewat ke KEPUTUSAN / ATURAN / memori.
- Commit di dalam
asisten/(git -C asisten add -A && git -C asisten commit -m "accio <tanggal>"). Kalau git belum punya nama/email, lewati commit dan bilang sekali caranya. - Balas: "Tersimpan. Berikut: [satu baris]."
Cara kerja di mode asisten
- Jawaban pendek dan tegas. Nol basa-basi, nol "Pertanyaan bagus!". Maksimal satu pertanyaan per balasan, bawa rekomendasi.
- "ya" / "gas" / "go" / "lanjut" = kerjakan, jangan ulang rencana.
- Empat skill dasar di
.claude/skills/aktif: tulis tanpa bau AI, cek alat/data dulu, bukti sebelum bilang selesai, tanya seperlunya. - Nol kirim keluar (email, WA, posting, pembayaran) dan nol push tanpa perintah eksplisit pemilik untuk tindakan itu.
- Jangan mengubah berkas kit (docs/, skills/, templates/, README, berkas ini) di mode asisten. Aturan pribadi masuk
asisten/ATURAN.md, supaya kit bisa diperbarui dengangit pulltanpa bentrok.
BELOW: example CLAUDE.md content showing real production patterns — the kit's own working rules, used as a live example of a disciplined CLAUDE.md.
CLAUDE.md — Claude Code Kit (this repo)
Repo type: Open-source methodology + templates + tooling. Audience: Solo founders + SEA/Indonesian developers + Claude Code adopters wanting opinionated infrastructure layer. Maintainer: Ruby Perkasa.
Hard rules when working in this repo
Voice + audience
- README + public-facing docs: English primary (audience global), but use Indonesian examples authentically (gue/lo in founder-style snippets is a feature, not a bug).
- Internal docs (this CLAUDE.md, plan files): Indonesian casual OK with maintainer, English fine for technical names.
- Do NOT translate Indonesian voice rules into English in templates — that destroys the differentiator.
- User-facing copy in templates uses universal
aku/kamu(warm-curator), NOT Jakarta slang (gue/lo). Ruby-Claude register ≠ Brand-User register.
Substance > polish
- Every doc must include a real example from a production venture (anonymized if needed). Abstract patterns without example = useless for learners.
- Templates must be FORKABLE, not theoretical. Test by trying to fork and run.
- Don't add a new pattern unless extracted from at least 2 different ventures (single-venture patterns risk being incidental).
Honesty
- This kit is NOT first-mover at most pillars. Owen Zanzal's Virtual Monorepo Pattern + claude-mem (46k+ stars) + WORKSPACE.md patterns predate parts of this. Acknowledge.
- Honest moat = composite + execution discipline + SEA-specific. Don't oversell.
- If a pattern competes with existing tooling (e.g., claude-mem), say so explicitly. Don't pretend they don't exist.
Anti-patterns to avoid
- "Vibe coding to riches" framing — this kit is the OPPOSITE thesis (infrastructure first, AI second)
- Toy demos or hello-world content (audience already past that)
- Generic "AI is amazing" filler
- Closed-source dependencies in the recommended path (Infisical OK, Doppler not — break OS positioning)
Working session protocol
Read sequence at start kit
README.md— positioning + current stateCLAUDE.md(this file) — rulesdocs/INDEX.mdif exists — what's shipped vs in-progress
Update protocol
- New methodology pattern extracted →
docs/##-pattern-name.md(numbered for read order) - New template →
templates/[name].template.md(always with.template.infix) - New example →
examples/[venture-anonymous]/[topic].md - New tool →
tools/[tool-name]/with own README
Commit hygiene
- Commits should be substantive. "wip" commits OK during private dev, squash before merging to public.
- Use conventional commits:
docs: add CLAUDE.md pattern walkthrough,feat(tools): add supabase usage watch action, etc. - Don't auto-push (per Ruby's holding-wide rule). Only push when explicitly asked.
Source material
Patterns come from the maintainer's own running businesses (booking software, travel marketplace, rental platform, content production, a strategic-memory repo that ties them together). Do NOT name the businesses, list local folder paths, or mention which venture a story came from. Keep the structure and the lesson; drop the identity.
A pattern needs n=2+: seen in at least two different projects before it becomes a rule here.
Versioning
- v0.0.x — bootstrap, no public release yet
- v0.1.0 — first useful release: README + 3 core docs + CLAUDE.md template + RUBY_BRIEF template
- v0.2+ — tooling additions (sync script, supabase guardrails, cost watch Action)
Do NOT include in this repo
- Real Supabase project IDs, service role keys, API tokens (even in examples — sanitize)
- Customer data, booking data, revenue figures
- Internal strategic decisions about other founders / partners
- Actual
RUBY_BRIEF.mdfiles from production repos (only.template.mdversions) - VPS IP / SSH keys / CDN or DNS API tokens
- WhatsApp numbers, real phone numbers, real emails (use
example.comdomains)
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

