tier-review
Skill1 starsChanged 4 months agoApache-2.0View on GitHub
Klasifikasikan perubahan kode (diff/PR/output subagent) ke Review Tier 1-4 dan jalankan prosedur review yang sesuai — Tier 1 line-by-line wajib untuk auth/session, RLS policy, multi-tenancy boundary, money calculation, audit-log write path, state machine, migration SQL, kalkulator business-critical, endpoint security-sensitive. Gunakan SEBELUM promote/commit/merge hasil subagent atau AI lain, saat user minta "review diff", "review PR", "cek hasil agent", "boleh di-merge?", atau setelah eksekutor melaporkan selesai.
In this repository9 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- adrSkill
- autonomous-runSkill
- docs-first-specSkill
- milestone-wrapupSkill
- session-endSkill
- session-startSkill
- vertical-sliceSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 65
Audit log: write path tidak bisa dilewati oleh jalur kode alternatif · tidak mencatat PII mentah/secret · append-only.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill 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
Tier Review — Kode AI-Generated
Prinsip: yang wajib adalah review diff + bukti verifikasi, bukan siapa yang mengetik. Tidak ada kode masuk branch utama tanpa melewati tier yang tepat. Tidak ada tool (/code-review, ultrareview, CI) yang menggantikan Tier 1 — semuanya pelengkap.
Step 0 — Syarat reviewer
Tier 1 direview oleh agen berkapasitas ≥ eksekutor (Opus me-review diff Sonnet). Kalau kamu adalah penulis kode ini di sesi yang sama dan diff-nya Tier 1 → nyatakan itu, dan minta review oleh sesi/agen terpisah tanpa konteks penulisan. Penulis me-review tulisannya sendiri mewarisi blind spot yang sama yang menghasilkan bug-nya.
Step 1 — Kumpulkan diff + bukti
- Diff lengkap:
git diff main...{branch}atau staged changes. - Bukti verifikasi dari eksekutor: output test (bukan klaim "test pass"), log run, screenshot bila UI. Tidak ada bukti = review belum bisa dimulai. Minta bukti dulu.
Step 2 — Klasifikasi per file/hunk
Ambil tier TERTINGGI yang menyentuh file tersebut:
| Tier | Cakupan | Treatment |
| 1 — Line-by-line wajib | Auth/session · RLS policy · multi-tenancy boundary · money calculation · audit-log write path · state machine transitions · migration SQL · kalkulator business-critical (voting/pricing/tax) · endpoint security-sensitive | Review penuh baris-per-baris + checklist di bawah. Temuan diringkas untuk dibaca user. Tidak ada pengecualian. |
| 2 — Sample | UI component · route non-sensitif · helper · test non-business-logic | Review 30-50% file berubah; fokus pola, bukan tiap baris. Frekuensi boleh turun bila pattern stabil. |
| 3 — Automated only | Build, TS strict, lint, test pass, migration valid | CI menegakkan. Cukup konfirmasi CI hijau. |
| 4 — Persona khusus | Security review pasca-foundation · legal/domain review kalkulator regulatif · UX review pasca primary flow · DevOps review pra-production | Terjadwal per milestone, bukan reaktif. Rekomendasikan bila milestone-nya tiba. |
Ragu antara dua tier → ambil yang lebih tinggi.
Step 3 — Checklist Tier 1 per kategori
Auth/session: password hashing (argon2/bcrypt) · token expiry + rotation · session fixation · logout membunuh session di server · rate limit di endpoint auth.
RLS / multi-tenancy: policy aktif per tabel tenant · tenant_id tidak bisa disuntik
dari client · query path yang bypass RLS (service role) diberi justifikasi · IDOR:
user A mengganti ID milik user B di URL/body → HARUS gagal.
Money: bigint sen + currency code, tidak pernah float · rounding eksplisit dan konsisten · operasi tidak overflow · pembagian (split/prorate) menjaga total.
Migration SQL: reversible atau punya recovery path · operasi berisiko (DROP/RENAME/ ALTER TYPE) pakai expand-contract · tidak menghapus data ber-retensi · ada catatan backup-sebelum-migrate.
State machine: transisi ilegal ditolak di server (bukan hanya disembunyikan di UI) · race pada transisi ganda · state persist konsisten dengan audit log.
Audit log: write path tidak bisa dilewati oleh jalur kode alternatif · tidak mencatat PII mentah/secret · append-only.
Authz umum (paling sering bocor di kode AI): check eksplisit per endpoint — "tidak ada UI ke sana" bukan proteksi · validasi input server-side (Zod) di tiap mutation.
Step 4 — Red flags lintas tier
Install library tak diminta (cek eksistensi + reputasi — slopsquatting) · perubahan
arsitektur tanpa justifikasi · orphan code · any / @ts-ignore tanpa komentar ·
TODO/FIXME menumpuk · klaim sukses tanpa verifikasi dijalankan · kepatuhan semu pada spec
(yang mudah diukur dikerjakan, constraint sulit diukur dilewati) · hardcoded
secret/API key · hardcoded business/legal rules.
Step 5 — Verdict
Format laporan:
## Review {branch/PR} — {tanggal}
| File | Tier | Temuan |
|---|---|---|
**Temuan** (per item): [BLOCKER|MAJOR|MINOR] {file:baris} — {masalah} → {perbaikan}
**Bukti verifikasi diperiksa**: {test output/screenshot/log — atau "TIDAK ADA" = blocker}
**Verdict**: APPROVE | APPROVE-WITH-NITS | BLOCK (alasan)BLOCK bila: ada temuan blocker · bukti verifikasi tidak disetor · diff Tier 1 tanpa test untuk skenario kritisnya. Setelah fix, review ulang bagian yang berubah — bukan rubber-stamp.
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.

