AGENT SKILL

language-code-review-edho-ferdian

Stack-specific review lenses on top of code-review: idioms, framework misconfigurations, ORM query traps, performance and test conventions across ~20 stacks. Load it for any review of code in a named framework, even one or two files. Use for "review kode Go/Python/React ini", "audit Django models", "cek FastAPI endpoint ini".

License 许可
MIT
所属仓库
edhoferdian/EEF
最近更新
2026-10-30
Stars
★ 22
Forks
0

01 前言 Front Matter

name
language-code-review-edho-ferdian
metadata
{ "last_reviewed": "2026-10-09" }
description
Stack-specific review lenses on top of code-review: idioms, framework misconfigurations, ORM query traps, performance and test conventions across ~20 stacks. Load it for any review of code in a named framework, even one or two files. Use for "review kode Go/Python/React ini", "audit Django models", "cek FastAPI endpoint ini".

02 资源文件 25

03 SKILL.md 正文

Language Code Review — Edho Ferdian Mode (Lens Layer)

You are still the same senior engineer from code-review-edho-ferdian — this skill does not replace that review, it sharpens it. A generic checklist catches generic bugs. It does not know that a Django serializer with fields = '__all__' leaks columns, that a FastAPI route awaiting nothing inside async def blocks the event loop, or that key={index} silently corrupts React list state. That is what this skill adds: per-stack idiom, framework-misconfiguration, and ORM/query knowledge, expressed as extra criteria slotted into the domains the general skill already reports under.

Relationship contract (read this first)

This is a lens layer, never a standalone review. It does not have its own Phase 0–5 pipeline, its own report format, or its own Reflection/Critique gate — it borrows all of that from code-review-edho-ferdian.

  • If code-review-edho-ferdian's Phases aren't already running, run them first. This skill's job is Phase 0's "conditional-lens detection" step and Phase 1's domain checklists — nothing else.
  • Every lens finding lands inside Domain 1 (CQ), Domain 2 (SEC), or Domain 3 (PERF) of the general report, using that domain's existing severity table, confidence labels, and finding template. A lens never opens a new top-level domain (contrast with accessibility-lens.md's A11Y-##, which the general skill already treats as a deliberate exception because accessibility doesn't fit CQ/SEC/PERF cleanly — language lenses do fit, so they stay inside those three).
  • Security items (SEC-08) no longer live in these reference files. Each references/*.md file below used to hold its own stack-specific security criteria inline; those have all moved to security-review-edho-ferdian/ references/language-specific.md as the single source of truth (removes the drift risk of the same criterion existing in two places). Each reference file still says, at its "Code placement" line, exactly which security items moved and where — load that file (or delegate to security-review-edho-ferdian directly) when Domain 2 needs stack-specific security depth.
  • Never duplicate what the general checklist already owns. Generic injection (SQL/command/path), generic secret handling, generic function length/nesting/magic-number checks, and generic N+1 detection are owned by references/review-checklist.md in the general skill. A language lens adds only what is specific to that language/framework — e.g. Django's select_related/prefetch_related mechanics for N+1, not N+1 itself. Every lens file below states this boundary again at its own top.
  • Both Phases 3 (Reflection) and 4 (Critique-Correction Loop) in the general skill already run over the whole finding set, lens findings included — this skill does not add a second reflection pass.

Phase 0 extension — stack detection

Run this as part of the general skill's Phase 0 "conditional-lens detection" step, immediately after the database/accessibility/RAG lens checks. Detect by manifest file first, extension second — a .py file alone doesn't tell you whether it's plain Python, Django, or FastAPI, but manage.py does.

Signal found Load Requires (load first)
package.json dependencies (dependencies or devDependencies) include react or react-dom references/react.md —
manage.py at repo root, or a settings.py with INSTALLED_APPS/django. imports references/python-django.md python.md
A FastAPI import (from fastapi import FastAPI / import fastapi) in main.py, app/main.py, or the file(s) in review scope references/python-fastapi.md python.md
celery in requirements*.txt/pyproject.toml, or a celery.py/tasks.py pattern in scope (add-on to the Django/Python detection) references/python-django-celery.md python.md + python-django.md
@nestjs/core/@nestjs/common in package.json, a nest-cli.json at the project root, or @Module/@Controller/@Injectable decorators in scope (detect per-project in an Nx/monorepo layout) references/nestjs.md —
package.json dependencies include @angular/core (load nestjs.md too when @nestjs/core is also present — e.g. an Nx monorepo with an Angular app and a Nest API) references/angular.md —
go.mod at repo root, or any .go file in scope references/go.md —
Cargo.toml at repo root, or any .rs file in scope references/rust.md —
package.json dependencies include vue (its own §9 sub-section covers Nuxt when nuxt is also present) references/vue.md —
composer.json has laravel/framework references/laravel.md —
pom.xml/build.gradle* has a spring-boot dependency, or @SpringBootApplication present (Quarkus sub-section inside the same file when quarkus dependencies are present instead) references/java-spring.md —
any .kt/.kts file in scope, or build.gradle.kts references/kotlin.md —
Package.swift, or any .xcodeproj/.xcworkspace references/swift.md —
package.json has react-native references/react-native.md react.md
pubspec.yaml has a flutter dependency references/flutter.md —
AndroidManifest.xml present, or a Gradle module applying the Android plugin references/android.md —
build.gradle.kts has org.jetbrains.compose references/compose-multiplatform.md android.md
any .csproj/.fsproj/.sln file references/dotnet.md —
CMakeLists.txt, or any .cpp/.hpp/.cc file in scope references/cpp.md —
torch import or dependency in scope references/pytorch.md code-review-edho-ferdian/references/mle-lens.md
any .pl/.pm/.t file, or cpanfile/Makefile.PL/.perlcriticrc at repo root references/perl.md —
oh-package.json5/module.json5 at repo root, or .ets files in scope references/arkts.md —
Gemfile at repo root, config/routes.rb, or .rb/.rake/.erb files in scope references/ruby.md —
Any .py file in scope and no Python framework row above matched references/python.md alone —

Multiple signals can be true at once — load every reference that matches (e.g. a Django project with a React frontend in the same repo loads react.md + python.md + python-django.md; a Django project that also uses Celery loads python.md + python-django.md + python-django- celery.md). State which lens file(s) you loaded in the Phase 0 summary, same as the other conditional lenses. If no signal matches (a stack without a reference file yet), skip silently — the general four-domain review still applies in full; there is just no extra lens on top yet.

A file in the "Requires" column assumes its prerequisite's general idiom checks already ran and only adds framework-specific criteria on top — each such file declares "Requires: ... (load first)" at its own top too.

Ground-truth-first rule

Every reference file below has a Ground-truth commands section with real, runnable commands for that stack. Run them — via the general skill's Phase 2 verification step — before asserting a lens finding as fact. This mirrors the general skill's own rule (database-lens.md already sets this precedent for EXPLAIN ANALYZE): reading code and guessing what a linter or type-checker "would" say is reasoning, not verification.

Unverified findings are capped at [Medium confidence]. A lens finding that could be confirmed by a ground-truth command but wasn't (tool not installed, not reachable, or you chose not to run it) never gets [High confidence] — say what command would confirm it, exactly as the general skill's Phase 2 confidence rule already requires.

Confidence floor & noise control

  • >80% confidence threshold before flagging. If you are not at least 80% sure a pattern is a real problem in this codebase's actual context (not just "this pattern is often bad"), don't flag it — or flag it at [Low confidence] with the specific uncertainty named, not silently omit the caveat.
  • Consolidate repeated findings. The same anti-pattern hit five times in one file (e.g. five Model.objects.all() loops missing select_related) is one finding with a count and all five locations listed, not five separate findings. This keeps the report actionable instead of noisy.
  • Never flag CRITICAL on unchanged/pre-existing code. If the review scope is a diff (the general skill's default) and the offending line existed before this change, cap it at MEDIUM and note it's pre-existing — CRITICAL is reserved for what this change introduces or what blocks this change from shipping safely. Pre-existing debt is real but it is not this PR's fault.

Reflection gate addition

Before finalizing any lens finding, add one more check to the general skill's Phase 3 Reflection pass: does this project's own configuration or convention already endorse this pattern? Check, in order:

  1. CLAUDE.md / AGENTS.md / .cursorrules at the repo root for an explicit statement that accepts the pattern (e.g. "we intentionally use fields = '__all__' on internal-only admin serializers").
  2. The stack's own linter/formatter config (.eslintrc*, ruff.toml / pyproject.toml [tool.ruff], .flake8) — a rule explicitly disabled with a comment explaining why is a documented exception, not a miss.
  3. A code comment at the exact site explaining the deliberate choice.

If any of these explicitly endorse the pattern, do not flag it — or flag it at INFO with "project convention, not re-litigating" rather than as a defect. This is the same false-positive discipline the general skill's false-positive-catalogue.md already applies; this is its stack-specific extension.

Lens maturity

Angular, NestJS, Python, and React lenses are field-proven; every other stack's lens is FOLD-M — medium-depth and plausible, but not yet verified against a real project, so its findings get the same confidence discipline as the Ground-truth rule above, not field-proven trust. Kotlin and the four mobile stacks (React Native, Flutter, Android, Compose Multiplatform) have no stack-specific section in security-review-edho-ferdian yet — route their security findings to the general SEC-01/02/10 codes and name the gap rather than inventing stack-specific criteria. Per-stack history, status, and provenance live in references/lens-status.md — open it when adding a stack or auditing where a lens came from, not on a normal review.

Language routing (inherited — see code-review-edho-ferdian, which points to skill-authoring-edho-ferdian's canonical contract)

Inherited, not restated — this lens has no report format of its own (see "Relationship contract" above), so it follows whichever language routing code-review-edho-ferdian is running under (itself pointing to skill-authoring-edho-ferdian §7). Nothing to configure here.

Global rules

  1. Lens, not a second review. Always runs inside code-review-edho-ferdian's phases, never standalone.
  2. No duplication. Generic injection/secrets/nesting/magic-numbers/N+1 stay owned by the general checklist; a lens adds only stack-specific depth.
  3. Detect by manifest, not extension, and load every matching reference — stacks can combine.
  4. Ground-truth or Medium-cap. Unverified findings never reach High.
  5. >80% confidence, consolidate repeats, never CRITICAL on pre-existing code.
  6. Check project convention before flagging — an explicitly endorsed pattern is not a defect.
  7. Provenance line in every reference file, stated once here.
  8. A FOLD-M lens is not field-proven — see "Lens maturity" above; treat its findings with the same confidence discipline as rule 4.

04 同仓库更多 Skills

edhoferdian
edhoferdian/EEF

video-style-edho-ferdian

Measure a reference video's look and pacing (colour grade as a 3D LUT, cut rhythm, overlays) and reproduce it: grade, cut and composite new footage, verify numerically, hand off an NLE timeline. Use for "tiru look video ini", "samakan grading", "bikin video dengan gaya referensi", "LUT dari referensi", "cut rhythm".

★ 22
edhoferdian
edhoferdian/EEF

test-authoring-edho-ferdian

Write unit and component tests well: React Testing Library, pytest, Go and Vue, plus regression-test patterns. Use for "tulis test untuk component ini", "bagaimana test hook ini", "test yang bagus untuk fitur X", "tulis test pytest/Go/Vue", or dev-kickoff's TEST stage.

★ 22
edhoferdian
edhoferdian/EEF

system-design-edho-ferdian

Mid-project architecture decisions: ADRs, trade-off analysis, non-functional requirements, scaling tiers and blast radius. Use for "desain arsitektur", "keputusan teknis besar", "bikin ADR", "trade-off antara X dan Y", "microservices atau monolith", "apa yang rusak kalau X diubah", or a capacity question.

★ 22
edhoferdian
edhoferdian/EEF

spec-mining-edho-ferdian

Extract behavioural specs from a codebase with none: WHEN/THEN requirements and invariants, each anchored to the code that enforces it, unclear behaviour recorded as uncertainty. Use for "ekstrak spec", "buat spec dari kode", "reverse-engineer the spec", "repo ini tidak ada dokumentasi", or a missing BEHAVIOR_SPEC in dev-kickoff.

★ 22
edhoferdian
edhoferdian/EEF

skill-authoring-edho-ferdian

Create and govern this ecosystem's skills: search before building, a quality bar, compliance checks, promoting repeated principles to rules, and packaging dist/*.skill. Use for "bikin skill baru", "ada skill buat X gak", "fork skill ini", "skill gue kepake gak sih", "package skill ini", "buatkan .skill-nya".

★ 22
edhoferdian
edhoferdian/EEF

skill-audit-edho-ferdian

Audit this repo's skills/ for staleness, redundancy, broken cross-references and weak descriptions (not the ~/.claude setup: that is config-hygiene). Use for "audit skill saya", "cek skill yang sudah dibuat", "ada yang redundan gak", "skill mana yang basi".

★ 22