Files
3x-ui/CLAUDE.md
T
Sanaei 5321665d5b feat(ci): give the review bot a severity scale and a tally
REVIEW.md said what blocks and what does not, but never how to mark a finding,
so every review invented its own shape and none carried a severity. It now
names the three markers the hosted Code Review service uses - Important, Nit,
Pre-existing - and keys them to what the pull request did rather than to how
alarming a defect looks alone: a defect it introduces or worsens is Important,
one it merely brought into view is Pre-existing and cannot be a reason to hold
it. Pre-existing was missing entirely, and checking what this panel emits means
reading far outside the diff, so those findings had nowhere to go except a
wrong Important or silence.

The volume cap said how many and never which. It now collapses a nit repeated
across files into one finding, prefers a nit in code the pull request wrote
over one in code it only moved, caps pre-existing findings at three, and states
that Important findings are never capped - a section listing two caps otherwise
reads as licence to trim what matters. The review opens with a tally so the
author sees the shape before the detail.

Two contradictions went with it. The file told the reviewer to skip what CI
enforces and then to check that a new i18n key reaches all 13 locales, which
i18n-dead-keys.test.ts pins in both directions - the rule moves to "Do not
report" with the reason. "Anything CI already enforces: npm audit" overstated
what runs; CI audits production dependencies at high and above, so a
dev-dependency advisory is out of scope by design.

The reviewer could not read its own CI. Only postgres-durable-first runs
against PostgreSQL, and XRAY_E2E_BINARY and XUI_SCALE_TEST are set by no job,
so a dialect or migration change can carry a wall of green while the paths it
touches never executed. That belongs to the verification bar, next to the rule
that a behaviour claim needs a file:line citation, and "CI passed" now needs a
run actually read. Also names the two house choices no linter defends: neither
golangci-lint nor oxlint rejects a testify or Tailwind import.

Both kinds of claim rot on a rename, so a test pins them the way
repo-context.md's claims are already pinned - the CI jobs REVIEW.md names must
exist in ci.yml, the skip gates it calls unset must stay unset, and the locale
count must match the directory.

The review itself moves from high to max effort, and the prompt records why it
names REVIEW.md at all: the code-review skill reads CLAUDE.md on its own but
not REVIEW.md, so dropping that clause would silently stop the file applying.
Drops a CLAUDE.md reference to tools/seedperf/, which no longer exists - the
review reads that file as project context, so a stale path there misleads it.
2026-08-21 03:37:12 +02:00

164 lines
10 KiB
Markdown

# CLAUDE.md
Operational guide for AI agents working in this repo. Long-form human docs:
`CONTRIBUTING.md` (setup, testing philosophy) and `frontend/README.md`.
Read those before large changes. This file is the short, must-follow version.
For a deep navigation map (request lifecycle, cron-job table, symptom → file
index, layering rules), read `docs/architecture.md` on demand — do not guess
file locations when it can answer in one hop.
## Stack
- Backend: Go 1.26 (`module github.com/mhsanaei/3x-ui/v3`), Gin, GORM.
Runs Xray-core as a managed child process (`internal/xray/process.go`) and
imports `github.com/xtls/xray-core` for config types + gRPC stats/handler/router
API. MTProto inbounds run a second managed child — the `mtg-multi` binary
(a multi-secret mtg fork — NOT a Go dependency; its prebuilt release binary is
fetched at image/release build time by `DockerInit.sh` + `release.yml`,
panel-side code in `internal/mtproto/`) — outside Xray, one process per inbound
serving each
client's FakeTLS secret via the fork's `[secrets]` section (plus per-client
ad-tags via `[secret-ad-tags]` and per-client data quota / expiry via
`[secret-limits]`, mapped from the client's `totalGB`/`expiryTime`). Client,
ad-tag and quota/expiry edits are hot-applied through the fork's management API
(`PUT /secrets`, bearer-token guarded) so connections survive; the manager
falls back to a process restart on older binaries. A client's panel-side
traffic reset also calls `POST /secrets/{name}/reset-quota` so a renewed client
is not re-blocked by the sidecar's quota counter.
- Storage: SQLite by default (`/etc/x-ui/x-ui.db` on Linux; the executable dir on
Windows), PostgreSQL optional (`XUI_DB_TYPE` / `XUI_DB_DSN`). The CGo SQLite
driver (`mattn/go-sqlite3`) needs a C compiler — `CGO_ENABLED=0` builds fail.
- Frontend: React 19 + Ant Design 6 + Vite 8 + TypeScript in `frontend/`,
built into `internal/web/dist/` (gitignored) and embedded via `embed.FS`.
## Repo map
- `main.go` — entry point + `x-ui` CLI (run, migrate, migrate-db, setting, cert).
- `internal/config/` — env parsing (XUI_DEBUG, XUI_LOG_LEVEL, XUI_LOG_FOLDER,
XUI_BIN_FOLDER, XUI_SKIP_HSTS, XUI_PORT, XUI_DB_*).
- `internal/database/` + `internal/database/model/` — GORM schema (~24 models;
Inbound, Client, Setting, User are the core), inbound Protocol enum,
AutoMigrate + hand-written migrations in `db.go`.
- `internal/xray/` — Xray child-process lifecycle, config generation, gRPC API.
- `internal/xray/geodata/` — streaming geosite/geoip `.dat` reader (cached
category index + paged entries) and `geosite:`/`geoip:`/`ext:` token parsing.
- `internal/mtproto/` — MTProto inbounds via the bundled `mtg-multi` binary.
- `internal/sub/` — subscription server (raw / JSON / Clash).
- `internal/eventbus/` — in-process pub/sub (outbound/node health, xray.crash,
cpu.high, memory.high, login.attempt).
- `internal/logger/`, `internal/util/` (link, crypto, sys, ldap, …),
`internal/tunnelmonitor/` — shared infrastructure.
- `internal/web/` — Gin server (embeds `dist/` + `translation/`).
- `controller/` — panel + REST API handlers; OpenAPI at /panel/api/openapi.json.
- `service/` — business logic (InboundService, SettingService, XrayService,
node sync); subpackages tgbot/, email/, outbound/, panel/, integration/.
- `job/` — 17 cron jobs (traffic, fail2ban IP-limit, node heartbeat/sync, LDAP,
CPU/memory watchdogs, …); full table in `docs/architecture.md` §5.4.
- `middleware/`, `entity/`, `global/`, `session/` (CSRF), `network/`,
`runtime/` (master/sub-node over mTLS), `websocket/`.
- `locale/` + `translation/` — i18n, 13 embedded locale JSON files.
- `frontend/` — React + TS source (see `frontend/CLAUDE.md`).
- `tools/openapigen/` — Go generator that emits frontend types + Zod/JSON schemas
into `frontend/src/generated/` from Go structs. The OpenAPI doc itself
(`frontend/public/openapi.json`) is assembled from those + `endpoints.ts` by
`frontend/scripts/build-openapi.mjs`.
- `docs/` — separate Next.js/Fumadocs site (pnpm, own CI in `docs-ci.yml`,
outside `make verify`). Holds a THIRD independent implementation of
link/subscription generation in `docs/lib/xray/` — check it whenever
share-link or install-command output changes.
## Hard rules (non-negotiable)
- Fix size must match bug size. Find the root cause, then make the SMALLEST
change that removes it — a one-line guard beats a new subsystem. A small bug
does not earn new columns, jobs, abstractions, config knobs or helper layers.
If a fix genuinely needs new architecture, say so and get agreement first;
never ship it unasked next to the fix.
- Comments in committed Go/TS: 2 lines MAX per comment block. Make the name
carry the meaning first and rename rather than annotate; spend the 2 lines on
the *why* a name cannot hold — an invariant, an issue number, a non-obvious
constraint. Exempt: `//go:build`, `//go:generate`, and other directives.
HTML `<!-- -->` is fine. (A linter cannot enforce this — you must.)
- New `g.POST`/`g.GET` in `internal/web/controller/` REQUIRES a matching entry
in `frontend/src/pages/api-docs/endpoints.ts`, then `make gen` (or
`cd frontend && npm run gen`). Hand-maintained but pinned both ways by
`TestRouteRegistryContract` (`internal/web/routes_contract_test.go`): a missing
OR stale entry fails `make test-go`. Scope: `/panel/api/*` + a few session
routes; sub-server routes are exempt.
- Response examples come from Go struct `example:` tags via `tools/openapigen`
never hand-write them. A new struct must be added to openapigen's `StructAllow`
allowlist (`tools/openapigen/main.go`) or it is silently omitted from
schemas/examples (and `build-openapi.mjs` then fails on the missing schema).
- A new or renamed endpoint has a FOURTH step nothing checks: copy
`frontend/public/openapi.json``docs/public/openapi.json`, then
`cd docs && pnpm gen:api` to refresh the MDX under
`docs/content/docs/en/reference/api/`. `docs-ci.yml` fires only on `docs/**`.
- A new English i18n key goes in EVERY locale JSON in `internal/web/translation/`
(13 files) AND must be referenced from `frontend/src` or Go in the SAME commit —
`frontend/src/test/i18n-dead-keys.test.ts` fails both ways. It is a frontend
test, so run `npm test`, not just `make test-go`. At runtime the frontend falls
back to en-US; Go (`internal/web/locale/`) returns "" for an unknown key.
- DB / model changes require a migration in `internal/database/db.go`.
- Every state-changing inbound/client op dispatches through `runtime.Runtime`
(`internal/web/runtime/`) — never straight to `internal/xray/api.go`, never from
a controller or cron job. A direct call passes every local test and silently
breaks every multi-node deployment. Other layering rules: `docs/architecture.md` §8.
- Conventional commits: `type(area): short imperative summary`, then a body
explaining the why. Types in use: `fix`, `feat`, `chore`, `refactor`, `perf`,
`docs`, `style`.
## Go conventions
- Stdlib `testing` only (no testify). Table-driven, `t.Run` subtests,
`t.Helper()` on helpers. Assert the exact value / typed error / emitted
string, never just `err != nil`. Prefer real deps over mocks: throwaway DB via
`database.InitDB(filepath.Join(t.TempDir(), "x-ui.db"))` +
`t.Cleanup(func() { _ = database.CloseDB() })`; `httptest` for HTTP.
`internal/sub`'s `initSubDB(t)` is the template.
- A test must fail without its fix. Write it, revert the fix, watch it go red,
restore. A test that passes either way is worse than no test: it certifies
nothing and then gets cited as proof the fix works.
- Test what can actually break. No test for a getter, a constant, a rename, a
pure map lookup, or inputs the function can never receive. One real test that
drives the bug through the actual code path beats five that restate the code.
- Code must pass `golangci-lint run` (gofumpt + goimports formatting): `make lint`.
- Postgres, xray-gRPC-e2e and scale tests `t.Skip` unless `XUI_TEST_PG_DSN`,
`XUI_DB_TYPE`+`XUI_DB_DSN`, `XRAY_E2E_BINARY` or `XUI_SCALE_TEST` is set — a
green `go test ./...` does not mean those paths ran.
## Frontend conventions (summary; full version in frontend/CLAUDE.md)
- Ant Design 6 only — no Tailwind/shadcn. Targeted tweaks, not rewrites.
- TS strict; oxlint's `typescript/no-explicit-any` is an error. Zod schemas in
`src/schemas/` are the source of truth; infer types with `z.infer`, never
hand-write. Do not edit `src/generated/`.
- Node 24 (`.nvmrc`) — `make gen` imports `.ts` directly and needs its type
stripping; Node 22 dies with `ERR_UNKNOWN_FILE_EXTENSION`. `npm test` includes
a headless-Chromium Storybook project, so run
`npx playwright install --with-deps chromium` once or `make verify` fails.
- Editing `frontend/src` does NOT change what users see until the Vite build is
regenerated into `internal/web/dist/`. In `XUI_DEBUG=true`, HTML is served from
the frozen embedded FS but JS/CSS off disk — after `npm run build` you MUST
restart `go run .` or you get a blank page with 404s.
- After touching share-link logic (`src/lib/xray/`), run `npm run test` (golden
fixtures); regenerate snapshots (`npx vitest run -u`) only for intentional
output changes, never to make a red test green.
## Build, test, verify
A fresh clone has no `internal/web/dist/`, so a bare `go build ./...` dies with
`pattern all:dist: no matching files found` while ~35 other packages pass — it
reads as a broken repo, not a missing step. Run `make dist-stub` once; every
`make` Go target already depends on it, which is why `make test-go` beats
`go test ./...`. Run `make help` for all targets. The local gate:
make verify # gen-check + lint + format-check + typecheck + test + build
# + build-storybook
That is the *fast* gate, not all of CI. `ci.yml` also runs `make race`,
`make vulncheck`, a live-Postgres job (where a SKIP counts as a failure) and a
30s fuzz smoke on `FuzzParseLink`/`FuzzDecodeCertPin` — run those locally when
you touch DB/dialect or parser code.
Common targets: `make gen` (regenerate Zod/OpenAPI), `make lint` (Go + frontend),
`make test` (Go `-shuffle=on` + frontend), `make race`, `make build`. See `Makefile`.
## Definition of done (before opening a PR)
1. `make verify` passes — its `gen-check` already runs `make gen` and fails on a
dirty `frontend/src/generated` / `frontend/public/openapi.json`.
2. Diff is focused; refactors are separate from feature work.