Shards edit items and add new ones, carrying cliloc ids no stock client table
has. The cliloc table now reads a SET of sources rather than one file — a base
plus every operator-maintained overlay under `custom/`, hash-gated together and
re-read on every boot, exactly as §6 reads the ServUO tree. Matching docs for
website #115.
website/CLILOCS.md gains a "Shard-added and shard-edited items" section: the
directory layout, merge precedence, the per-source breakdown an operator uses
to confirm an overlay took effect, and why `custom/` is a convention we chose
rather than one discovered from the shard — ServUO has no server-side notion of
a custom cliloc, they live in the patched client a shard distributes and
nothing in the tree declares them. Scale recorded: the live shard's script tree
references 16,434 cliloc ids and only 37 are absent from stock, which is why
this is an overlay and not a second table.
"Why there is no staged-approval flow" is replaced by "Two ways a refresh is
refused", because the set brings back the hazard a single file did not have. A
corrupt source fails the parse loudly; a source that has VANISHED parses
perfectly and imports a table quietly missing everything it contributed. That
is the same ambiguity §6 stages a facet removal for, so it is staged here too
(`needsReview`, `{approve:true}` to accept) — as a flag rather than §6's
approve/reject pair, because the atlas stores a pending decision SO THAT
approving re-parses, and here nothing is stored.
Two more traps recorded in §8.6 (now five), both found by running a shard-style
overlay rather than another stock-table fixture:
- Tidying punctuation unconditionally corrupts real names — a custom
"Runic Gateway Sigil (v2)" rendered as "(v2". Stripping leftover brackets is
right after a placeholder is removed and wrong otherwise, the same condition
the `%` rule already had.
- Source labels must be forward-slashed and root-relative or the same directory
fingerprints differently on Windows and Linux and every boot looks like a
change — the identical bug §6 records.
BACKEND_DESIGN.md: the source set, the two refusals, and the updated admin
route notes (`approve`, `missingSources`, overlays found beside a file path).
Documentation only. Every claim was written after the behaviour was observed:
the counts come from the live shard's script tree, and the refusal/approve flow
and per-source breakdown are copied from real runs in website #115.
PROJECT_TREE.md files are deliberately untouched — CI-generated by the
sync-project-tree workflow.
Co-Authored-By: Claude <noreply@anthropic.com>
Protocol 3.0 §8.6 resolved as its own website-only change, landing ahead of
§8 so the marketplace ships with real item names. Matching documentation for
website #TBD.
NEW website/CLILOCS.md — operator-facing: why the conversion step exists, how
to convert, how to configure the path, the refresh contract, what gets stored
and how names are applied.
link/v3.md §8.6 rewritten. Two things in the original recommendation turned
out to be wrong, and both are recorded because the reasoning generalises:
1. The committed db/data/clilocs.json artifact predates the Part C
corrections (§6.1) and violates both — no committed snapshot of derived
content, and nothing EA-derived ever shipped. UO's strings are EA's,
exactly as the creature sprites are.
2. "scripts/buildClilocs.js reads the UO client's Cliloc.enu" is not
possible. EVERY current client ships its cliloc files compressed (first
DWORD's high byte 0x8E, the Mythic container); the plain layout is what
those files looked like before that change, and parsing one as the other
does not fail cleanly — it yields ~19k records with negative ids, 1,722
distinct keys out of 19,508, one 62 KB "string", and a truncation
somewhere in the middle. ServUO's own Ultima.StringList cannot read it
either, so VendorSearch.GetItemName is already inert on such a shard and
the work could not be pushed to the plugin.
That second point also retires an open question in §8.2: the warning never to
call GetItemName in the market sweep costs us nothing we could otherwise have
had, because the in-game Vendor Search gump has the same gap.
Three traps found by building it are recorded: StringList.SaveStringList
RE-COMPRESSES on save (its output is byte-identical to its compressed input,
because its purpose is round-tripping a file back into the client); trimming a
text line before splitting silently drops the ~half of a table that is empty
strings; and Number('') is 0, not NaN.
Also updated:
- Progress and §9 sequencing tables: order 5 split into 5a (this, website
only) and 5b (the four-repo wire change).
- website/BACKEND_DESIGN.md — shard_clilocs / shard_cliloc_meta, the three
admin routes, and why there is no staged-approval flow and no public route.
- link/INTEGRATION.md — the char.profile field note now says explicitly not to
expect the shard to resolve clilocs, and points at CLILOCS.md.
- §10 documentation obligations list CLILOCS.md.
Documentation only. Every claim was written after the corresponding behaviour
was observed running: the compressed-format finding and the parse failures
come from the real client files on this machine, and the counts (123,490
parsed → 67,496 stored) and timings from importing them into the live MariaDB.
PROJECT_TREE.md files are deliberately untouched — they are CI-generated by
the sync-project-tree workflow and say so in their header.
Co-Authored-By: Claude <noreply@anthropic.com>
Protocol 3.0 §7 lands across servuo-plugins, link and website; this is the
matching documentation.
INTEGRATION.md
- points.board in the event catalog: one frame per system, never a delta, no
points.remove (the shard's system set is fixed at startup). Four gotchas
called out, all of them things a consumer gets wrong by default:
1. maxPoints: 0 means UNCAPPED, not "zero allowed" — and on a real shard
it is the COMMON case, so anything rendering points/maxPoints must
special-case it or divide by zero on the happy path.
2. nameString is usually null, with a cliloc in nameNumber — so
humanising the system key is the primary display path, not a fallback.
3. players counts players actually holding points, not table size: ten of
the ~25 systems keep a zero-point row per character ever created, so
the raw count would report the shard's whole census.
4. Entries carry serial + name only, never acct/webId.
- The char.profile `points` block, and why `rank` is absent by default.
- GET /points and /points/:system, including why 404 (never published) and
200-with-empty-top (published, nobody scored) are different answers.
v3.md
- B/2 marked done in both the progress table and §9.
- NEW §7.5, "what the run against a real shard changed" — the same record §6.1
and §6.2 keep. Four corrections the plan could not have anticipated from
reading PointsSystem.cs, the sharpest being that (long)double.MaxValue is an
unchecked conversion yielding long.MinValue, which published
"maxPoints": -9223372036854775808 on the first live sweep. Also records that
GetEntry/GetPoints mutate the world on AutoAdd systems and so cannot be used
in a read model, and the one deliberate deviation from §7.4: the visibility
field rule must key on the wire's `name`, not the descriptive
`characterName`, or it is silently inert.
BACKEND_DESIGN.md — shard_points_boards (including why the top-N list stays in
the payload and why listing orders by COALESCE(name, system)), plus the two new
public routes.
PLAN.md — 3.0 phasing brought current: the spawn atlas and points.board added to
what has shipped, and the Points* keys noted in the config-key paragraph.
PROJECT_TREE.md files are deliberately untouched — they are CI-generated and say
so.
Co-Authored-By: Claude <noreply@anthropic.com>
Docs half of website #113 (Protocol 3.0 Part C, second website PR).
Carries the atlas rewrite that missed #67: that PR merged before the "derive
from the tree on every boot" commit was pushed, so `edge` currently describes
the build/import-artifact design that was rejected in review, not what shipped
in website #112. It lands here.
New in SPAWN_ATLAS.md: the six public routes and five admin ones, and three
behaviours that read as bugs unless they are written down — an unreadable tree
answers 200 with status "unavailable" rather than 500 (refresh reports outcomes
so boot is never blocked by a bad tree, and the contract is preserved at the
API), setting the ServUO path deliberately does not import, and `points` is a
count while `spawners` is the list.
Also the delay-unit trap: XmlSpawner stores MinDelay/MaxDelay in minutes OR
seconds per record, decided by that record's own DelayInSec flag, so a `5` is
five minutes on one spawner and five seconds on the next. Both are plausible
respawn times, which is what makes it silent. 170 of 6,455 stock spawners are
second-flagged. And PARSER_VERSION, which exists because hashing the tree alone
would strand an install whose maps never change on whatever an older parser
derived.
BACKEND_DESIGN.md gains the routes, the router-map entry, and the parser-version
rule. v3.md marks Part C done and records in 6.3 what the API half found.
api-route-inventory.json refreshed from the live manifest — it had drifted to
200 routes before this PR (real count was 204) and is now 215.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01U7CBg11prhLimL9iHSX1bP
Follows the redesign in website #112. Two decisions from the original §6 were
rejected in review and replaced; the docs now describe what was actually built.
**The committed artifact is gone.** A shard's maps change over its life, so a
snapshot in the repo silently drifts from the world players actually see. The
ServUO tree is the single source of truth and the atlas is re-derived on every
server boot, hash-gated so an unchanged tree costs one read pass and no write.
**Nothing may name a facet.** The first implementation carried a lookup table of
the six stock UO facets. A shard may add facets, replace them outright, or rename
them when its maps are updated, and a built-in list mishandles all three
silently. Reconciliation is now by matching against the facet set discovered from
the shard's own data, with an unmatched name keeping its own rather than being
forced into a wrong bucket.
## Changes
- **`website/SPAWN_ATLAS.md`** rewritten: the two ideas that shape the design,
how to configure the tree path, the boot flow as a decision tree, the
approve/reject flow, and the code layout. The artwork policy is unchanged and
still explicit — no art ever ships, operators extract their own from their own
client files.
- **`link/v3.md` §6.1 (new)** records the two rejected decisions plus the two
boot-path contracts. The old "what real data changed" list becomes §6.2. §6's
now-superseded passages — the artifact bullet, the payload budget, the operator
re-run story — are marked rather than deleted, so the reasoning stays legible.
- **`website/BACKEND_DESIGN.md`** documents `shard_atlas_pending` and the two
contracts that make it safe: a facet removal is staged for a human, and the
boot refresh can never block startup.
The two contracts are the part worth reviewing. Losing a facet is
indistinguishable at boot from a half-copied or mid-update tree, so it is staged
rather than applied; and no failure mode of the atlas — missing path, unreadable
mount, malformed file, database error — is allowed to stop the site coming up.
---
- [x] AI-assisted: written with **Claude Code** (Claude Opus 5), reviewed before opening.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01U7CBg11prhLimL9iHSX1bP
Protocol 3.0 order 3 (Part C), docs half of website #112. Part C is website-only
— no plugin, no sidecar, no new kinds, no wire change.
## New: website/SPAWN_ATLAS.md
The operator-facing reference: the build/import split and why it exists (build
needs a ServUO tree, import does not, and the container has the artifact but not
the tree), the re-run story, the artifact format, the placement transform, and
the three quirks in the source data that are silent when unhandled.
Also documents the artwork policy explicitly: **the project ships no creature
art and no extraction tooling.** Sprites live in the operator's own client
.mul/.uop files and are theirs, not ours to redistribute. `art` is nullable and
NULL on every fresh import; an operator who wants art extracts it themselves into
the gitignored uploads/atlas/ and maps slugs in a gitignored art map. Text-only
is the normal, supported state — not a degraded one.
## New: v3.md §6.1 — what the build against real data changed
Six corrections, kept as a diff rather than edited into §6 in place, because
each is a trap the next person would otherwise re-enter:
1. **Six facets, not thirteen.** Eodon.xml and the other named-area files carry
TerMur/Trammel points; the facet comes from each record's `<Map>`.
2. **The XML dependency call resolved: hand-rolled, zero deps.** §6 left
fast-xml-parser vs a tokenizer open.
3. **Facet names disagree between sources** — Locations says `Ter Mur`, `<Map>`
says `TerMur`. Unreconciled the landmark fallback never fires there and every
unregioned Ter Mur/Tokuno spawn silently reads "Wilderness".
4. **Spawn type tokens carry XmlSpawner directives** (`Fairy,{RND,4,8}`,
`alchemist/z/-50`). Taken literally they invent creatures that do not exist
and split real ones in two. 71 of 845 affected; 800 remain after stripping.
5. **The artifact is 1.41 MB, not "well under 1 MB"** — down from 4.40 MB via
three encodings. Getting under 1 MB would mean dropping the spawner name.
6. **DELETE, not TRUNCATE** — TRUNCATE is DDL in MariaDB and implicitly commits,
which would defeat the all-or-nothing reload the design asked for.
§6 also now records that Part C ships as two website PRs: the parsing half is
where the correctness risk lives and should not be reviewed inside a 10k-line
diff alongside routes and React.
## BACKEND_DESIGN.md
The seven atlas tables, the import-owned contract, the four column choices that
are traps (`spawn_range`/`grp` reserved words, DELETE vs TRUNCATE, explicit point
ids, plain INDEX not FULLTEXT), and the distinction between the configured
champion roster and the live champ.update feed.
PROJECT_TREE.md is left alone — it is auto-generated by the sync-project-tree
workflow.
---
- [x] AI-assisted: written with **Claude Code** (Claude Opus 5), reviewed before opening.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01U7CBg11prhLimL9iHSX1bP
Protocol 3.0 order 2 (v3.md §5) is built across all four repos; this is its
documentation half, plus the running progress record the plan was missing.
v3.md
- A progress table at the top and a State column on §9's sequencing table, so
"what has landed" is answerable without reading four git logs. Part A (order
1) and world.ruleset (order 2) are marked done; the spawn atlas is next.
- §5 gains the implementation notes worth keeping, chiefly: where a system's
on/off state is DERIVED rather than configured, read the system's own static
instead of inventing a .cfg key (Shadowguard has no Enabled key — it's the
TOL expansion gate; Factions is `!ViceVsVirtueSystem.Enabled` by
construction in stock ServUO). Also that the plugin CAN be compile-verified
despite the "no standalone build" caveat, and how.
INTEGRATION.md
- The world.ruleset catalog entry and GET /ruleset, with the two things
consumers get wrong: caps are in TENTHS (1000 = 100.0), and `connect` exists
only if the operator set Bridge.PublicConnectAddress — the shard's real
listen address is never published.
- §2 now says plainly that v3 has NOT been bumped yet and what that means:
sidecars on `edge` report 2 while already carrying some v3 kinds, so do not
infer feature availability from the version during this window.
PROTOCOL_2.md §10.4
- The deferred "which PvP system does this shard run?" is answered (VvV on,
Factions off — and mutually exclusive by construction), and world.systems is
marked superseded by world.ruleset, which carries the systems block it asked
for. No orphan kind is left behind.
BACKEND_DESIGN.md — the shard_ruleset table (why it is stored whole rather than
normalized, and why no row means null rather than {}) and the public route.
PROJECT_TREE.md is deliberately untouched: sync-project-tree regenerates it on
push to main, so it updates itself at the v3 cutover.
Co-Authored-By: Claude <noreply@anthropic.com>
The live five-rung smoke test of the visibility framework found that Part
A enforced it on the SSE path and on /guilds + /governors, but not on the
remaining public REST reads - so one event was projected live and served
verbatim from stored history.
link/v3.md gains 3.6.1 with the full list (the anonymous acct/webId leak
on /feed, the flattened ownerAcct on /idoc, the dead `houses` field
rules, /feed ignoring live config, the empty-allowlist fall-through, and
the Date-to-{} projection bug), plus the rule it leaves behind: a read
path that returns shard data and does not project is a bug, and every new
Part B/C surface must gate its kind set on live config rather than on
PUBLIC_KINDS.
3.5 also corrected: the table is NOT seeded on boot. An absent row means
"use the compiled default", which keeps the defaults in one place instead
of duplicating them into a seeder that could drift.
BACKEND_DESIGN.md 6.5 records the same as a security contract: rule 1
locks a field by meaning rather than spelling; PUBLIC_KINDS is a
module-load constant and must not answer per-caller questions;
projectFeature walks arrays and plain objects only.
SHARD_VISIBILITY.md gets the admin-facing version - that stored history
answers the same way the live stream does, and that turning live updates
off stops the push, not the reading.
Co-Authored-By: Claude <noreply@anthropic.com>
Protocol 3.0 Part A. Admin-configurable, per-feature and per-field
audience control over every shard-derived surface, replacing the static
PUBLIC_KINDS allowlist that used to be the whole boundary.
- SHARD_VISIBILITY.md (new): the admin-facing guide - the ladder, what
each of the ten features exposes, the defaults, the two rules that are
code rather than configuration, and worked examples.
- BACKEND_DESIGN.md 6.5 (new): the same thing as a security contract -
the ladder and how viewerLevel resolves it, the locked acct/webId rule,
the fail-closed kind map, the asymmetric ladder fallbacks, and the
three enforcement points. Plus the shard_feature_visibility schema, the
/public/shard/features route, and the adminOnly tier on
/admin/shard/visibility.
Defaults reproduce pre-3.0 behavior everywhere, with one deliberate
exception which is the leak Part A was written to close: guilds and
governors previously returned the raw stored payload, whose leader and
governor actors carry acct and webId, to anonymous callers.
Co-Authored-By: Claude <noreply@anthropic.com>
Doc side of website + Android-app feat/sso-trusted-device.
TRUSTED_DEVICES_MFA.md §6 gains an "SSO login paths" subsection: SSO is not
exempt from the second factor, and a trusted device skips it exactly as on the
password path (previously SSO consulted trust nowhere, so an external-identity
user was asked for a code on every sign-in). Documents the callback-side skip,
the new trustDevice/deviceName on POST /auth/sso/totp, and why recovery codes
stay password-login only.
Also writes down how this reaches the Android app, since it is not obvious: the
app's SSO runs in a Custom Tab that shares the system browser's cookie jar, so
the rg_trust cookie covers native SSO with no app change and no trust token in a
start URL (which would leak a secret into query strings and logs). The app's own
token is minted at /auth/mobile/sso/exchange instead — an authenticated
app→server call — so it never travels in the deep link, and the bridge row holds
only a boolean. Notes that one tick yields two independently-revocable rows.
§4 documents the new mobile_auth_sessions.trust_device column; BACKEND_DESIGN.md
gets the same column in its bridge table, the trust note on the /exchange row,
and a pointer from the bridge intro to the Custom Tab cookie model.
Co-Authored-By: Claude <noreply@anthropic.com>
Two corrections found by a live smoke test of all 200 routes at every access
level (website PR: fix/uolink-client-throw-and-sitemode-gate).
ARCHITECTURE.md: the "uoLinkClient never throws" invariant was true of the HTTP
call but not of resolving the config, which decrypts the stored auth token and
throws when the ciphertext can't be authenticated (SECRET_ENC_KEY rotated, or a
DB dump restored under a different key). Spell out that this is now handled
inside the client, reported as { ok: false, error: 'uo-link config unreadable' }
with a distinct ERROR log, and that GET /admin/uo-link/config keeps working —
it is the screen an admin needs to re-enter the token and recover.
BACKEND_DESIGN.md: GET /dashboard is staff-wide while PUT /site-mode on the
same screen is adminOnly — the one place a single screen spans two tiers. Note
that the client must gate that control itself rather than relying on the route
gate that admitted the user to the page.
Co-Authored-By: Claude <noreply@anthropic.com>
The domain split is complete. API_V2_PLAN.md gains a "PR 5 — as landed"
section (route table, the four zero-diff gates, and the findings worth
carrying forward) and its status line and sequencing list are updated: only
the CSP enforce PR remains, blocked on soak data rather than on code.
BACKEND_DESIGN.md §2 replaces the auth.routes.js / public.routes.js entries
with the full per-capability tree for auth/, public/ and player/, and §4's
group headings now point at the index.js files. The /player prose names the
three routers behind the shared gate.
Findings recorded rather than left in the code alone:
- public/ and auth/ deliberately have no group gate — the obvious hardening
edit to either is an outage.
- GET /auth/me depends on session.router.js being mounted last, because
use('/me', meRouter) matches the bare /me and supplies its noindex header.
- Two root-mounted routers (public/site, auth/session) on the PR 4 dashboard
precedent, safe only because neither declares router-level middleware.
- loginGuards is the PR's shared module, the counterpart to PR 3's
imageUpload.js.
- Filename deviations from the target tree (posts not news, session.router.js
added) and why public.controller.js was not split.
Co-Authored-By: Claude <noreply@anthropic.com>
Covers website PR 4, the last admin split PR: shard (16), uo-link (5), email
(6), discord-bot (2), settings (2) and dashboard/site-mode (2) leave the
residual file, which is deleted. The admin group is fully split.
API_V2_PLAN.md gains a "PR 4 — as landed" section recording the two decisions a
reviewer would otherwise have to reconstruct: dashboard.router.js is mounted at
the group root (the single relaxation of the mount-at-a-prefix rule, safe only
because it declares no router-level middleware), and /shard keeps two gate tiers
in one router because prefix ownership beats swagger-tag grouping. Sequencing
item 7 is marked landed; PR 5 (public/player/auth) is the only split PR left.
BACKEND_DESIGN.md §2 gets the six new routers in the folder tree and drops the
residual entry; §4's /admin preamble now describes the ops/config gates instead
of pointing at a file that no longer exists.
WIKI_UPGRADE.md's two links into admin.routes.js are repointed at wiki.router.js
and admin/imageUpload.js.
Co-Authored-By: Claude <noreply@anthropic.com>
Adds a "PR 3 — as landed" section to API_V2_PLAN.md and ticks the sequencing
list. 31 routes extracted, 33 left in admin.routes.js; all four zero-diff gates
came back clean and 434 server tests passed.
Findings carried forward:
- The residual 33 is exactly PR 4's list, so admin.routes.js is deleted by
PR 4 rather than PR 5.
- First shared module in the split: the multer config, because POST
/posts/upload and POST /uploads no longer live in the same file.
- POST /uploads keeps its Admin · Posts swagger tag — retagging is a real
OpenAPI diff and does not belong in a route-move PR.
- The wiki router has load-bearing intra-file route order (/categories and
/tags ahead of /:slug) that no gate can catch, because the manifest sorts
its entries. Verified by introspecting the built router stack instead.
BACKEND_DESIGN.md §2 gets the four new routers plus imageUpload.js in the
folder tree, and §4 notes that the content capabilities add no gate beyond
staffOnly.
Co-Authored-By: Claude <noreply@anthropic.com>
Matches the code change in website: 18 more admin routes carved into
moderation.router.js (15), botActivity.router.js (2) and activity.router.js (1),
leaving 64 in the residual admin.routes.js.
API_V2_PLAN.md gains a "PR 2 — as landed" section recording the four zero-diff
gates and two decisions worth carrying into PRs 3-5:
- /activity gets its own file rather than the target tree's plan to park it as
a singleton inside dashboard.router.js — honouring the tree would have left
one route in the residual file for two PRs, and it is a genuinely separate
capability (the staff audit log, not the dashboard's stats overview and not
the botScore middleware's ban state). PR 4 therefore mounts dashboard and
site-mode only; the target tree is updated to match.
- A gate moves to a router-level `use` only where it was already a *prefix*
mount (moderation's modAccess). Bot-activity's per-route adminOnly stays
per-route, because the per-route handler count is the only thing in
routes.guards.json that would catch a dropped gate — requireRole(...) returns
an anonymous arrow and never appears by name.
BACKEND_DESIGN.md §2 (folder structure) and §4 (the /admin contract preamble) are
updated for the new files and their gates. PROJECT_TREE.md is left alone — since
website#98 it is auto-generated by the sync-project-tree workflow.
Co-Authored-By: Claude <noreply@anthropic.com>
Documentation half of the first of five domain-split PRs (API_V2_PLAN.md § Phase 2).
BACKEND_DESIGN.md
- §2 folder structure: admin/ now shows index.js (shared gate + mount table) and
the four capability routers with their route counts, prefixes and extra gates;
admin.routes.js is labelled as the 82-route residual that goes away with PR 5.
- §4 /admin heading: was "admin.routes.js -> admin.controller.js", now points at
admin/index.js and notes staffOnly, which the old heading omitted.
- The "planned change" note becomes "in progress" with what has landed.
API_V2_PLAN.md
- Status: planning -> in progress; PR 1 marked landed in the sequencing list.
- New "PR 1 — as landed" section: the route-count table (6+15+3+4+82 = 110) and
three findings for PRs 2-5 — why the self-service /shard/* routes stay with the
shard capability despite their Admin · Account tag, why a prefix mount must not
be "simplified" to a pathless one (a bare use(gate) would then run for requests
headed to later mounts), and that routes.guards.json came back zero-diff too.
- New section on the swagger path-normalization prerequisite and its consequence:
with sorted path keys, a pure route move produces no spec diff, so the spec
becomes a third zero-diff gate alongside the manifest and guards files.
- Correction to step 6: PROJECT_TREE.md is auto-generated by the sync-project-tree
workflow since website#98 and must not be hand-edited in split PRs.
api-route-inventory.json is unchanged — verified still byte-identical to
server/routes.manifest.json (200 public + 2 internal), which is the point.
Co-Authored-By: Claude <noreply@anthropic.com>
Matches website PR "build(swagger): normalize and sort generated OpenAPI path
keys", which post-processes swagger-autogen's output ahead of the admin router
domain split (API_V2_PLAN.md § Phase 2).
- website-README.md § Regenerating the spec: why trailing slashes are stripped
(a capability router mounted at /users declaring router.get('/') would document
/api/v1/admin/users/, a URL no client calls) and why path keys are sorted.
- BACKEND_DESIGN.md § generated artifacts: note that both the route manifest and
the spec are emitted sorted, so a diff in either is proportional to the change.
Co-Authored-By: Claude <noreply@anthropic.com>
Companion to website "feat(security): soak the tightened CSP on report-only".
The plan's Phase 1 claimed a two-directive delta, one of which was adding
`form-action 'self'` as "currently absent". It was not absent. The directives
object in app.js does not list it, but the middleware runs `useDefaults: true`
and helmet's default set already supplies it, so production has been serving it
all along. The plan was written from the config rather than from the live
header; the correction, and how it was caught, are now recorded in place rather
than quietly fixed.
That leaves `frame-ancestors 'self'` -> `'none'` as the entire behavioural delta
of the phase. Worth noting that this is also the directive that most justifies a
soak: a frame-ancestors violation is reported by the browser of whoever framed
the site, so it is the only available way to discover a legitimate embed before
an enforcing policy breaks it.
Also documented:
* POST /api/csp-report -- the same-origin sink report-to/report-uri point at,
why it is same-origin, why it lives outside /api/v1, both wire formats, the
Reporting-Endpoints header requirement, and the properties that make an
unauthenticated public POST safe (always-204, caps, truncation, rate limit).
* That the sink is scoped to the soak, so the enforce PR must decide
explicitly whether to retire it or keep a report-to group on the enforced
policy -- rather than leaving an orphan route behind.
* The `[csp]` log tag in section 7.5 as the thing to watch during the soak,
and what silence across one release means.
* client/vite.config.js already sets `modulePreload: { polyfill: false }`, so
the plan's inline-polyfill prerequisite was already satisfied.
api-route-inventory.json moves 199 -> 200 for the new route. That is the PR 0
freeze working as intended: the first manifest diff since the baseline is a
deliberate, reviewed one.
Co-Authored-By: Claude <noreply@anthropic.com>
Companion to website PR 0 (chore(server): freeze the URL surface with a
generated route manifest).
BACKEND_DESIGN.md gains § 4.0, naming the two generated artifacts that are
actually authoritative about the API and what each is authoritative *for*: the
manifest records which URLs exist (introspection-derived, reality), the Swagger
spec records what they mean (annotation-derived, intent). The prose tables in
§ 4 are orientation and can drift; those two files cannot. Also documents
routes.guards.json as a review aid that is explicitly not a contract.
API_V2_PLAN.md marks PR 0 shipped and records its two deviations. The optional
unauthenticated-status snapshot was tried and dropped exactly as that section
allowed — against the dead-port mariadb pool the tests use it sits on the acquire
timeout rather than failing fast — replaced by a deterministic assertion that
every /admin/** and /player/** route still carries requireAuth.
routes.guards.json is committed and staleness-checked even though a diff in it
is not a contract change, because an ungenerated review aid rots into a
misleading one.
The sequencing section is corrected: PR 0 now runs before the CSP pair. The CSP
report-only PR must stand up a POST /api/csp-report collector for `report-to` to
target, which is a new URL under /api/**; landing it first would have left PR 0
generating 200 routes against a 199-route baseline, destroying its own acceptance
test. With PR 0 first, the collector appears as a reviewed, deliberate +1 in the
manifest — the mechanism working as intended.
api-route-inventory.json is unchanged, which is the point: the generator
reproduced it byte-for-byte on first run.
Co-Authored-By: Claude <noreply@anthropic.com>
The API v2 plan is revised down to the work that is actually justified: a CSP
hardening pass and an in-place domain split of the monolithic route wiring.
- Auth merge (httpOnly cookies -> bearer + rotating refresh for every client) is
removed and re-filed as deferred behind trigger conditions. httpOnly+SameSite
is the stronger model, session.service.js already unifies cookie and bearer,
the SSO/PKCE transaction cookies survive any merge, and it dragged the admin
SSE fetch/ReadableStream rewrite along as a dependency for no user-visible
payoff. A revival must first spec refresh-token reuse detection and a rollback
procedure.
- No parallel /api/v2. The URL surface is already grouped by capability, so each
new router file mounts at the prefix it already owns and every URL stays
byte-identical. No dual mount, no per-route migration, no v1 retirement; the
SPA, Discord bot, and Android app are all untouched. API_V2_SKELETON.md is
marked superseded (kept as the recipe if a versioned API is ever forced).
- The /api/mobile facade and app-version floor are deferred with the revival note
that it starts as a one-line alias mount, not ~70 hand-written delegates. The
M11 milestone is dropped from android/PLAN.md.
- Adds PR 0: a generated route manifest, so "every URL is unchanged" is proved by
a zero-line diff rather than asserted in review. The baseline
api-route-inventory.json (199 API routes + 2 internal) is committed here and is
what PR 0's generator must reproduce byte-for-byte.
- Split sequenced as five grouped PRs; CSP fixed to report-only first, then
enforce (the old plan contradicted itself), with the verified delta being just
form-action 'self' and frame-ancestors 'none'.
Co-Authored-By: Claude <noreply@anthropic.com>
Match the website change: the ntfy relay is reached through the public
reverse proxy, which runs outside the compose network and can only reach
a service via a published host port. Update the "no published host port /
internal-only publisher" claims in android/PLAN.md (§11 + §13) and
website/BACKEND_DESIGN.md to describe the published NTFY_HOST_PORT
(default 2586 -> ntfy:80), and note that both devices and the backend
publisher reach ntfy on the public origin.
Co-Authored-By: Claude <noreply@anthropic.com>
Record that the whole /player/* router sits behind requireAuth only (not
requireRole('player')): staff are a superset of players, every handler is
self-scoped to the caller, and staff reach the identical handlers under
/admin/shard/*. This is why a staff account with linked characters gets
its "My characters" and personal notification streams on the mobile
client. Matches the code change in RunicGateway/website.
Co-Authored-By: Claude <noreply@anthropic.com>
Add TRUSTED_DEVICES_MFA.md (the approved design/implementation plan) and fold
the feature into BACKEND_DESIGN §3 (trusted_devices + recovery_codes schema),
§4 (login/totp trust+recovery, /auth/me/trusted-devices*, recovery-codes*,
admin trusted-device + /mfa/reset routes), and §6 (trusted-device security
model + audit actions). Note the app-side trust/recovery flow in android PLAN §4.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Replace the vague "helmet with a CSP suited to the SPA" line with the actual
policy now implemented in server/src/app.js: per-directive sources and the
rationale for each non-'self' allowance (Google Fonts, inline React styles,
external/embedded images, same-origin REST+SSE), why upgrade-insecure-requests
is omitted, the scoped looser CSP for the /api/docs Swagger UI route, and the
X-Powered-By handling across the public and internal listeners.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NgyHnrNa8WwG3doxvxjuCr
Promote APP_LINKS.md from a deferred design note into an implementation spec
matching the website `feat/mobile-app-links` and android `feat/app-links`
branches: the server-side `/.well-known/assetlinks.json` route + `mobile_app_links_enabled`
toggle + additive redirect-allowlist entry, and the app-side `autoVerify`
intent-filter driven by a build-time `appLinkHost` (a single multi-tenant APK
cannot autoVerify open-ended shard domains, so App Links are a white-label /
first-party build opt-in; the custom scheme stays the permanent fallback).
Update PLAN.md §9 (M9 follow-up) and the §14 open item, and the redirect-URI
allowlist section of website/BACKEND_DESIGN.md.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NgyHnrNa8WwG3doxvxjuCr
Record the plan before coding: native Android "Sign in with Google/Discord"
via a Mobile SSO Authorization Bridge that extends the existing /auth/sso/*
redirect flow and terminates in the existing mobile bearer tokens.
- BACKEND_DESIGN.md: mobile_auth_sessions / mobile_auth_codes schema, the
/auth/mobile/sso/{start,exchange} contract, the two PKCE layers, state/CSRF,
exact-match redirect-URI allowlist, TOTP parity, and the documented
revocation-latency window.
- android/PLAN.md: promote §4.2's "possible later enhancement" to milestone M9
(backend-first, mirroring M7); status note.
- android/APP_LINKS.md: new architecture note on the per-shard assetlinks.json
/ pairing multi-tenancy question (App Links deferred; custom scheme only now).
Co-Authored-By: Claude <noreply@anthropic.com>
Flip M7 Part 2 to landed (Android-app#15) and record the two implementation
decisions: the direct-ntfy embedded distributor (no UnifiedPush library — the
plan's stated likely path; foreground-service SSE, no second app, no Google
Play Services), and the small additive push.ntfyUrl settings field the app
needs to discover the relay (website#79). Document push.ntfyUrl +
NTFY_PUBLIC_URL in BACKEND_DESIGN.md.
Co-Authored-By: Claude <noreply@anthropic.com>
- BACKEND_DESIGN.md: push_devices + notification_subscriptions tables (§3), the
/auth/me/devices* + /auth/me/notifications/* API rows (§4), a push-notification
design + security section (content-free tickles, PUBLIC_KINDS split, owner-keyed
personal streams, SSRF endpoint guard, untrusted-relay model), and the ntfy
compose service in the deploy section (§8).
- PLAN.md: flip M7 Part 1 (backend + docs) to in-review — status line, §8 item 3,
§9 M7.
Co-Authored-By: Claude <noreply@anthropic.com>
Counterpart to RunicGateway/website#77.
- BACKEND_DESIGN.md: /public table now documents the new GET /public/version
(DB-free identity/version), the version block on /public/status, and the brand
block on /public/settings (per-shard theming: name/accent/logo/hero/favicon).
- android/PLAN.md: mark §8 item 4 (version/health) and item 6 (branding) DONE and
update the prerequisite-progress summary — all v1 prerequisites are now done;
only push notifications (item 3) remains and is post-v1 (M7). App M0–M4 unblocked.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NgyHnrNa8WwG3doxvxjuCr
Counterpart to RunicGateway/website#76 (role-agnostic /auth/me/* self surface).
- BACKEND_DESIGN.md: add the /auth/me/account* rows to the /auth API contract and
a note that the surface reuses account.controller behind requireAuth (any role),
so a client manages its own account without touching /admin.
- android/PLAN.md: mark §8 item 1 (role-agnostic self-service) DONE and update the
prerequisite-progress summary; version/health (item 4) and branding (item 6)
remain open, push (item 3) is post-v1.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NgyHnrNa8WwG3doxvxjuCr
Add /auth/password/forgot and /auth/password/reset/:token to the API
contract and the password_resets table to the schema section, matching the
website change (RunicGateway/website feat/password-reset). Notes the
no-enumeration behaviour, single-use hashed-token model, and that the Android
app hands off to the web reset page (PLAN.md §4.2).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NgyHnrNa8WwG3doxvxjuCr
Retire env-var SMTP basic-auth and send the contact form through Gmail over
OAuth2 (SMTP XOAUTH2), configured in Admin -> Settings -> Email via an in-app
"Connect Gmail" consent flow. Reuses the existing google SSO OAuth client; the
captured refresh token is stored AES-GCM-encrypted (write-only over the API,
never returned), mirroring the auth-provider and Discord-bot secret patterns.
- schema: new email_config singleton table (mirrors bot_config)
- model: emailConfig.{db,model} with encrypted refresh token + getSafe/getWithSecret
- mailer: nodemailer OAuth2 transport (client id/secret from the google provider
row), contact recipient = contact_email setting, mailto: fallback preserved,
plus sendTest()
- routes/controller: /admin/email config, connect start+callback (ssoState CSRF
+ PKCE), test, disconnect
- client: EmailDelivery section on the Settings page + api methods; Settings copy
now spells out that contact_email is the delivery recipient
- docs/env: drop SMTP_*/CONTACT_TO from env examples; update README/BACKEND_DESIGN
- tests: emailConfig.model + mailer suites (8 new; full suite 142 pass)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XKeCQEJZr1AFJN4Bgcmvh3
