a24fcf0f26b629166faf3d9de0d1966aa3916c8d
11 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 sulthan
|
1552dd15da |
fix(docker): correct the timezone claim — UTC is the tell, not a country mismatch
The previous commit documented the clock-zone requirement as "the zone must match the egress IP's country". Re-measuring against the deployment case shows that is wrong. The original inference came from reading the host's /etc/timezone (Asia/Bangkok) and assuming the egress IP was Thai. It is not: this host egresses from an Indonesian IP. Asia/Bangkok cleared the challenge not because it matched a country but because it is simply not UTC, and the two share +07, which hid the distinction. Measured 2026-08-08, identical container, one Indonesian egress IP: TZ=UTC never cleared (60s, twice) TZ=Asia/Jakarta cleared in 4s TZ=America/New_York cleared in 4s America/New_York matches neither the country nor the offset nor the hemisphere, and clears just as fast. So a UTC clock is itself the bot signal - Cloudflare scores it as the datacenter default - and any real zone satisfies the check. This makes the knob considerably less fragile than documented: BROWSER_TZ needs a plausible zone, not a geolocated one, and a deployment that moves region does not have to keep it in sync. Comments in chrome/entrypoint.sh and docker-compose.yml, the hard constraint in AGENTS.md, and the PR description are corrected accordingly. BROWSER_TZ is also documented in .env.example for the first time, which is the file an operator actually copies - the setting decides whether kagane works at all, and a UTC server (the common case) is exactly the one that fails with it unset. Re-verified against the shipped image with TZ=Asia/Jakarta: TestSmokeKaganeImage PASS (5.00s, 56710 bytes of image/webp), TestSmokeKaganeGet PASS (1.24s, status 200). |
||
|
sulthan
|
5713e3d04a |
fix(docker): replace headless-shell with real Chrome so kagane's challenge clears
chromedp/headless-shell cannot clear kagane.to's managed challenge. It is
a stripped Chrome build, and the tells are structural rather than a
header: navigator.webdriver is true, the plugin list is empty, and the
client hints are Chromium- rather than Chrome-branded. Overriding
webdriver through CDP was tried on its own and changed nothing.
Everything below was measured on 2026-08-08 from a single IP, against the
same kagane cover, so the comparisons are like for like:
chromedp/headless-shell:stable never cleared (90s)
zenika/alpine-chrome never cleared - ships Chrome 124, old
enough that Cloudflare refuses it and
old enough to break chromedp's CDP structs
google-chrome, default UA never cleared (60s) - --headless=new
advertises "HeadlessChrome"
google-chrome, stock UA, UTC never cleared (90s)
google-chrome, stock UA, TZ set cleared in ~4s
So both remaining tells are load-bearing, and each was tested in
isolation. chrome/ is a Debian image with google-chrome-stable, a UA
whose version is read back out of the binary at startup (a hardcoded one
would drift out of step with the Sec-CH-UA hints on the next Chrome
update and become a fresh tell), and no --enable-automation.
The timezone matters because Cloudflare scores a browser whose clock zone
disagrees with its egress IP's country as a proxy. Note that the usual
`-v /etc/localtime:/etc/localtime:ro` does not work here: Chrome resolves
the zone through ICU, which takes the name from that path's symlink
target and ignores the file's contents, so glibc reports the host zone
while Chrome still reports UTC. /etc/timezone carries the name and is
mounted instead; BROWSER_TZ overrides it for a host whose clock is UTC in
a country that is not.
Chrome also binds its DevTools port to loopback and silently ignores
--remote-debugging-address, which is why headless-shell fronted it with
socat. This image does the same, so it stays a drop-in: the compose
service keeps the headless-shell name and its pinned address, and
BROWSER_WS_URL is unchanged.
Deploying needs `docker compose build headless-shell`.
|
||
|
sulthan
|
2ef769d421 |
Open registration to guild members (#27) (#36)
Closes #27. Guild membership is now the whole gate. `discordCallback` checks membership (and `DISCORD_REQUIRED_ROLE` when set), then `Store.EnsureReader` creates the Reader on first sight and returns the same row on every later login. The refusal returns before `EnsureReader`, so a turned-away sign-in leaves no row behind. `OWNER_DISCORD_ID` still seeds the owner, but only as the administrator — it no longer gates login. The cutover grace path goes with it: `API_TOKEN`, `API_TOKEN_GRACE_UNTIL` and the legacy branch in `httpmw.ResolveReader` are deleted, so a credential authenticates exactly one Reader or nothing. `userscript.Handler` drops its re-derivation too — the resolved path segment is already the credential. New surfaces: an empty library offers both install links (behind the tab-specific empty states, so "No favourites yet" still wins), and the owner alone gets a Readers panel with `POST /readers/{id}/revoke`. The owner's own row is not revocable — 404, not a self-logout. Isolation is asserted from both directions for read, modify and delete, and the shared-series invariant is pinned: two Readers on one series produce one series row, two independent progresses, one poll per due cycle, and one Reader's delete leaves the other's bookmark and the poll intact. Verified: `go test ./...` green; live smoke against a throwaway Postgres — empty-library state in both colour branches, roster rendering, a real revoke through the panel (target 401s next request, owner untouched), owner self-revoke refused 404, per-Reader `/u/<cred>` and bearer auth both 200 with 404 for an unknown credential. Reviewed-on: #36 Co-authored-by: Sulthan Zaki <sultankiki05@gmail.com> Co-committed-by: Sulthan Zaki <sultankiki05@gmail.com> |
||
|
sulthan
|
27cf0955de |
Per-Reader userscript credential with UI install and rotation (#24) (#32)
Closes #24. Child of #18; based on current main (includes Postgres, Reader table, Discord OAuth).
## What
Each Reader's userscript credential is derived from `TOKEN_KEY`, their Discord id and a token epoch (HMAC-SHA256, hex); only its SHA-256 sits in `readers.token_sha256` (new `token_epoch` column, migration 0006). One credential authenticates the script download path and the API bearer header.
- `internal/token`: derivation + hashing; the seed refreshes the owner's epoch-0 hash only before first rotation, so a restart can never resurrect a rotated-away credential
- `httpmw.Auth`/`ResolveReader`: acting Reader resolved from the credential hash, stashed in request context; the retired global `API_TOKEN` resolves to the owner until `API_TOKEN_GRACE_UNTIL` (enforced in code, logged per use) on both the bearer and script-download paths
- Userscript handler renders the bindmounted file with the resolved Reader's credential substituted for `__API_TOKEN__`; a legacy-path request during grace serves the derived credential, so installed devices self-migrate on their next update poll
- Web UI: "Userscripts" panel — session-gated install endpoints render the script directly (credential never in markup, address bar, or a redirect), confirm-gated rotation with an atomic epoch bump + hash rewrite and a reinstall warning
- Both userscripts carry `__API_TOKEN__` placeholders; the committed global-token literal is removed
## Design note
Credentials are derived rather than stored-random because the server must rebuild install URLs after restarts while the DB holds only hashes. HMAC output is high-entropy and unbrute-forceable; the AC's intent (unguessable, DB-leak-proof) is met.
## Deploy (also in DEPLOY.md)
1. Add `TOKEN_KEY` (`openssl rand -hex 32`) — required; changing it later invalidates every credential.
2. Keep `API_TOKEN` + set `API_TOKEN_GRACE_UNTIL` for the 14-day window.
3. After deploy, sign in → Userscripts → reinstall both scripts on every device. This also retires the old global credential for real — its literal survives in git history (present since
|
||
|
sulthan
|
bcc6b45515 |
feat(backend): Discord OAuth login with DB-backed sessions (#23) (#31)
Implements #23 per ADR-0002. - Discord authorization code grant (identify + guilds.members.read), form-encoded token exchange - Guild membership gate via the single-guild endpoint; optional DISCORD_REQUIRED_ROLE (empty default) - Owner Discord ID is the only identity allowed to sign in - Sessions are DB rows with opaque random ids; cookie carries only the id; expiry enforced; delete = revoke - HMAC session signing, derived key, and WEB_PASSWORD removed; no replacement signing secret - Login rate limiting preserved on the callback - Full flow tested through the real router against a local Discord stub (DISCORD_API_BASE) - Env: DISCORD_CLIENT_ID/_CLIENT_SECRET/_GUILD_ID/_REQUIRED_ROLE/_API_BASE/_REDIRECT_URI; docs updated go test ./... passes. Reviewed-on: #31 Co-authored-by: Sulthan Zaki <sultankiki05@gmail.com> Co-committed-by: Sulthan Zaki <sultankiki05@gmail.com> |
||
|
sulthan
|
08749df050 |
feat(backend)!: run on Postgres with a migration-owned schema (#28)
Swap modernc.org/sqlite for jackc/pgx/v5 with no observable change: same endpoints, same wire format, same updated_at ordering rule. The schema now comes from numbered SQL embedded in the binary and applied on startup, one transaction each, recorded in schema_migrations. That replaces two pieces of SQLite-era machinery, both deleted rather than ported: the column probing (Postgres has ADD COLUMN IF NOT EXISTS, and there is no legacy database left to probe) and the Asura key rewrite, which has run clean on every start for months now that the userscripts strip build hashes before writing. Its regexp survives as latest.asuraBuildHash, where the poller still needs it to scope chapter links to a series whose slug carries a rotating hash. Types get real: favorite is a boolean, chapter numbers double precision, timestamps stay unix-ms bigint. SQLite's null-safe IS NOT becomes IS DISTINCT FROM, which is what implements the rule that only reading progress reorders a list. Inside COALESCE/NULLIF the status and kind parameters need an explicit ::text -- there is no target column to infer from and Postgres refuses to guess. Tests lose their free t.TempDir() database, so Docker is now a hard prerequisite for `go test ./...`: internal/pgtest starts one postgres:17-alpine per test binary and hands each test a database of its own. Also lands CONTEXT.md and the four ADRs written while scoping #18. BREAKING CHANGE: DB_PATH is retired for DATABASE_URL, which is required and has no default. Compose gains a postgres service on an internal network with its own volume; POSTGRES_PASSWORD joins .env. The old bookmarks-data volume is deliberately left undeclared so `docker compose down -v` cannot take the pre-migration database with it. main is not deployable until #25 and #26 land. Closes #20 Co-authored-by: Sulthan Zaki <sultankiki05@gmail.com> Co-committed-by: Sulthan Zaki <sultankiki05@gmail.com> |
||
|
sulthan
|
b9f9aea82c |
docs: secure-coding rules for agents, and refresh stale AGENTS.md top-matter (#16)
Two doc commits: a new secure-coding rules section, plus a fix for top-matter the rebrand left stale.
## `9156525` — secure-coding rules
`AGENTS.md` carried two security invariants (bearer auth, CORS) but nothing about the code an agent actually writes here. That is the gap worth closing: measured rates for AI-generated web/backend code are ~40% vulnerable (Pearce et al.), 45% failing security tests (Veracode 2025), and users *with* assistants shipped SQLi at 36% vs 7% for the control group (Perry et al., Stanford). The failure classes cluster on broken access control, injection, session/error handling and invented dependencies — all live surfaces in this repo.
Rules were **extracted, not pasted**. Every one names a guard that already exists in-tree, so the instruction is *match this*, not *invent something*:
| Rule | Existing anchor |
| --- | --- |
| parameterized SQL only; constants may concatenate | `store.go` — all queries use `?` |
| `html/template` only; no `template.HTML` on stored data | `web.go:85` |
| client-supplied URLs pass the fetch gate | `poller.go:144 fetchableSeriesURL` |
| cap remote bodies | `fetch.go:16 maxBodyBytes` |
| `subtle.ConstantTimeCompare`, never `==` | `middleware.go:22`, `session.go:61` |
| generic error out, detail to log, never log the token | `web.go:151` |
| `X-Forwarded-Proto` for Secure; **rightmost** XFF for IP | `session.go:68,101` |
| cookie flags; expiry checked before signature | `session.go:72-94`, `Verify` |
| validate at handler boundary | `handlers.go:48` `MaxBytesReader` 64 KB, 400 on bad key/status/kind |
| site strings via `el({text})`, never `{html}` | `el()` in both userscripts |
| `fetch()`/`authHeaders()` → `API_BASE` only | existing `authHeaders` |
| `localStorage` = cache/queue, never credentials | shared with site JS |
Plus a dependency rule (stdlib first; verify a package exists before adding — ~20% of LLM-proposed packages don't resolve, which is the slopsquatting vector) and a review gate marking auth/CORS/session/crypto/fetch-gate as security-critical.
Deliberately **excluded**: container signing, k8s admission control, IaC scanning, PII/HIPAA/PCI, C/C++ memory safety. Per OpenSSF's guide for AI assistant instructions, irrelevant rules make a model generate code compensating for attacks that cannot happen. None of those apply to a single-user Go + SQLite + userscript stack.
Sources: OWASP AISVS 1.0 Appendix C, OWASP Top 10 / ASVS v5, OpenSSF *Security-Focused Guide for AI Code Assistant Instructions* (2025-08-01).
## `5d4d890` — stale top-matter
The rebrand rewrote root `AGENTS.md` as a compression pass and switched Bromite -> Violentmonkey, but left the project described as a manga-only tracker over two sites. Six sites, two libraries and two userscripts now exist.
Root `AGENTS.md`:
- *What this is* names both scripts with their site lists, the `kind` column, and the `<site>:<series_id>` key shape.
- Origins constraint generalised past Asura/Demonic.
- Records that **kagane and novelfull are reliably Cloudflare-challenged** and browser-polled over CDP. Without it that bullet list reads as contradicting the code, since the paragraph above asserts blocking is "not universal — and not reliably reproducible".
- Diagram says two userscripts.
`backend/AGENTS.md` — two instances of the same defect, found while verifying the above:
- Store key list gained `novelfull|lightnovelworld` and the `kind` column.
- **`NOVEL_USERSCRIPT_PATH` documented** — it shipped in `main.go:150` undocumented.
The dated Cloudflare paragraph is left verbatim: it is a timestamped observation ("Verified 2026-07-26"), so rewriting it would falsify a record rather than update it. `userscript/AGENTS.md` is untouched; it already documents both novel adapters and the `LIBRARY`/`STORE_PREFIX` split.
## Verification
Docs-only, no code touched. Every code reference above was read at `4229c17` before being cited — the fetch gate, body cap, constant-time compares, cookie flags, handler validation, `el()` helper and both route registrations. No invented line numbers.
## Not addressed here
The `API_TOKEN` literal is committed in plaintext in both userscripts and in their `@downloadURL`/`@updateURL` lines. The new rules say not to propagate it, but the actual remedy is rotation plus build-time substitution, since the value is already in git history. Separate change; flagging it so it does not get lost.
Reviewed-on: #16
Co-authored-by: Sulthan Zaki <sultankiki05@gmail.com>
Co-committed-by: Sulthan Zaki <sultankiki05@gmail.com>
|
||
|
sulthan
|
4229c179b0 |
rebrand: MangaBM → BookmarkManager, add novel library support (#15)
Two intertwined changes — the rebrand and the novel library were developed on
the same branch because the novel UI plumbing is part of the new "Bookmark
Manager" wordmark in the web shell.
## What it does
- **Rebrand**: MangaBM → BookmarkManager across the Go module, compose stack,
env vars, Traefik hostnames, container/image names, userscript storage
prefixes (`mangabm:cache` → `bmgr:manga:cache`, `mangabm:queue` → `bmgr:manga:queue`),
and docs.
- **Novel library**: same backend, two libraries. New `kind` column splits
bookmarks into `manga` / `novel`; PUT validates it. Two userscripts:
- `manga-bookmark.user.js` — unchanged behaviour, just stamps its own `kind`.
- `novel-bookmark.user.js` — separate Violentmonkey install with adapters
for **novelfull.com** (polled via headless browser — Cloudflare JS
challenge) and **lightnovelworld.net** (polled via plain TLS).
- **Web UI**: library switch on the app shell. Login art, libswitch, and
novel-site colours from the Cinder design snapshot.
## Plumbing
- `addedColumns` ALTER for `kind` runs on first start after upgrade; every
pre-existing row is backfilled to `'manga'`. No manual SQL, no down-time.
- `ALLOWED_ORIGINS` gains the two novel sites.
- New `NOVEL_USERSCRIPT_PATH` env (default `/userscript/novel-bookmark.user.js`),
bindmounted alongside the manga script.
- Traefik router names `mangabm*` → `bmapi*` / `bmweb*`.
## Test status
- `go test ./...` — green
- `node --test userscript/test/logic.test.js` — 34 pass
- `node --test userscript/test/novel-logic.test.js` — 11 pass
- `node --check` on both userscripts — clean
## Notes for the redeploy
.env keys were renamed (`MANGA_API_HOST` → `BOOKMARK_API_HOST`,
`MANGA_WEB_HOST` → `BOOKMARK_WEB_HOST`). Update DNS / Traefik labels on the
prod override before pulling, otherwise the public hostnames go dark.
See the redeploy instructions I'll post next to this PR.
Reviewed-on: #15
Co-authored-by: Sulthan Zaki <sultankiki05@gmail.com>
Co-committed-by: Sulthan Zaki <sultankiki05@gmail.com>
|
||
|
sulthan
|
180ee78b1f |
Add comix.to and kagane.to support (#13)
Tracks read progress on comix.to and kagane.to alongside asura and demonic, in both the userscript and the backend. Implements `docs/superpowers/plans/2026-08-03-comix-kagane-support.md`. ## Userscript - `comix` adapter — `/title/<id>-<slug>`; only the id prefix is identity (the slug follows the title). No `og:image`, so the cover is matched by `alt`. - `kagane` adapter — reader URLs are uuids with no chapter number, so it comes out of `og:title`; anchor scanning is structurally impossible, replaced by `latestChapterFromApi` against kagane's same-origin JSON API. - `seriesId` threaded through `latestChapterFromAnchors` so comix can scope its scan to its own series and a recommendation strip cannot win the maximum. - `@match` for both hosts, panel chips, v1.6.0. ## Backend - `latestChapterFrom` cases: comix parses the SSR JSON state blob (`latestChapterUrl`, scoped to the series id); kagane parses API JSON (`chapter_no`). - Poller allowlist extended; `Poller.BrowserFetch` with `fetcherFor(site)` routes kagane to a browser fetcher. Nil means kagane is not polled at all — never a fallback to the TLS fetcher, which would only ever retrieve a challenge page. - `BrowserFetcher`: chromedp against a `headless-shell` sidecar. kagane sits behind a Cloudflare JS challenge that no TLS fingerprint clears, and the request is made inside the page rather than by replaying `cf_clearance`. - `BROWSER_WS_URL` wiring, sidecar in both compose files (no `ports:`, dedicated non-external network), Dockerfile on `golang:1.26-alpine` — chromedp requires go 1.26. - Web UI `--comix` / `--kagane` tokens in both colour branches. ## Notes for review - `series_url` is client-supplied and a headless browser is a strong SSRF primitive, so kagane's host is pinned twice: in `fetchableSeriesURL` and again in `kaganeAPIURL`. - Three chained defects found during verification made the browser path dead under Compose (sidecar flag collision, Chrome's Host-header DNS-rebinding check, the wrong chromedp option). Fixed; the compose comments record the wrong configurations too, so they don't get "simplified" back. - `ALLOWED_ORIGINS` now includes both new origins. Without it every write from comix/kagane silently fails CORS preflight, parks in the retry queue, and drops at the cap. ## Verification 221 backend tests, 32 userscript tests, static `CGO_ENABLED=0` build, both compose configs. Two gaps, both real: 1. The userscript on live pages via Violentmonkey needs a human browser profile — not run. Check: comix series page (title/cover, no chapter), comix chapter page (records the number; an *older* chapter must not regress it), comix SPA navigation without reload, kagane series page (og:image cover), kagane reader (number from `og:title`), both chips opening the right sites. 2. The kagane browser path has not completed end-to-end anywhere. Dial/navigate/fetch is confirmed, but Cloudflare 403'd headless-shell's Chrome on every attempt from the dev sandbox, and comix's poll-through-Docker was blocked by that environment's TLS interception. Both environment-dependent rather than branch defects — the first real deploy is the actual verification. Reviewed-on: #13 Co-authored-by: Sulthan Zaki <sultankiki05@gmail.com> Co-committed-by: Sulthan Zaki <sultankiki05@gmail.com> |
||
|
sulthan
|
0416354c06 |
Serve the userscript from the backend; card + loading fixes (#8)
Serves the userscript from the backend so Violentmonkey auto-updates it, plus two panel fixes.
## Backend: `GET /u/{token}/manga-bookmark.user.js`
The script is read off disk per request from `USERSCRIPT_PATH` and streamed back with its `@version` line rewritten.
- **Token in the path, not a header.** Violentmonkey's update poll sends no `Authorization` header, and the script embeds `API_TOKEN` in plain text — an open URL would hand that token to anyone who guessed it. Compare is constant-time.
- **404, never 401**, for both a wrong token and a missing file: a prober learns nothing about whether the route exists.
- Registered outside `withAuth` and outside the `WEB_PASSWORD` gate, so the script is installable on a deployment that never enabled the web UI.
- Stdlib only (`crypto/subtle`, `os`, `regexp`) — no new Go dependencies.
**The served `@version` is derived from the file's mtime** (`YYYY.MM.DD.HHMM`, UTC), discarding whatever the file body says. Violentmonkey only updates when the served version sorts higher than the installed one, so a body-derived version means one typo or accidental downgrade freezes updates forever. An mtime-derived version is monotonic by construction. A file with no `@version` line is served byte-identical. `os.Stat` runs before `os.ReadFile`, so a concurrent edit can only serve new content under an old stamp — which self-heals on the next poll — never the reverse.
## Bindmount
`./userscript` is bindmounted read-only at `/userscript`. The script is deliberately **not** copied into the image: the build context stays `./backend`, and widening it would churn every `COPY` path for a file the mount always supplies. Editing the file on the VPS is live on the next poll — no rebuild, no restart. `git pull` restores the committed version, so a redeploy always ships the repo's script; checkout sets mtime to now, so even a rollback serves a *higher* version and is adopted. Without the mount the endpoint 404s and logs it; bookmark sync is unaffected.
`@downloadURL` / `@updateURL` are literal URLs in the metadata block — it is parsed before any JS runs, so `API_BASE`/`API_TOKEN` cannot be interpolated. The token was already committed in this file, so this adds no new exposure.
## Userscript UI
- **Card actions moved under the subtitle.** Only the cover and the title continue reading now; the subtitle and the action row are inert siblings in the text column. A thumb that misses ★ lands on nothing, and Remove is never inside a link.
- **Loading spinner** while the first fetch is in flight — the panel used to read as frozen on the first open after a cold start. It draws only when there is nothing cached to draw instead, so a populated list never flaps.
## Verification
- `go test -count=1 ./...` — ok, 7.070s
- `node --check` clean; `node --test userscript/test/logic.test.js` — 14/14
- Live `docker compose` smoke: `/healthz` 200, wrong token 404, script served with a stamped `@version 2026.07.28.1057` and both metadata URLs present; `touch`ing the file advanced the served version to `2026.07.28.1100` with no restart.
Layout and spinner are verified on-device — there is deliberately no DOM test harness.
Reviewed-on: #8
Co-authored-by: Sulthan Zaki <sultankiki05@gmail.com>
Co-committed-by: Sulthan Zaki <sultankiki05@gmail.com>
|
||
|
sulthan
|
206c447f36 |
Stable Asura series IDs: strip rotating build hash (#6)
Asura series slugs carry a site-wide build hash (-059befe1) that rotates on every redeploy, silently orphaning all asura bookmarks (old-hash URLs 302 to new-hash ones, so detect() yields keys that never match stored rows).
Backend:
- migrateAsuraKeys in OpenStore: one-off idempotent migration rewriting hashed asura keys to the stable hashless ID, merging collisions to newest updated_at; losers deleted before winner rewrite (PK-collision safe) — regression tests included
- poller latestChapterFrom: build hash made optional in chapter-scoping regex so latest_chapter survives redeploys
Userscript:
- stripBuildHash(/-[0-9a-f]{8}$/) applied to seriesId in both asura detect branches; URLs keep full slug (stale hashes 302)
- load-time migration rewrites cached + retry-queued asura keys to the stripped form so a queued PUT cannot resurrect an orphaned row
Docs: AGENTS.md + CLAUDE.md URL-shape notes; design spec at docs/superpowers/specs/2026-07-28-asura-stable-series-id-design.md
Verified: go test -count=1 ./... green (new rotation + collision tests), node --check green, regex verified against live asurascans.com slugs.
Deploy order: backend first (migration runs at OpenStore). Orphan rows created by not-yet-updated userscripts self-heal on next backend restart.
Reviewed-on: #6
Co-authored-by: Sulthan Zaki <sultankiki05@gmail.com>
Co-committed-by: Sulthan Zaki <sultankiki05@gmail.com>
|