19902a9f02429ecf9c1ab992b6a63f5288e6c46f
13
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
  adminandClaude Opus 5.5
|
48ec722660 |
Let an AI agent see the open editor through MCP (#429)
Claude Code should be able to work inside the user's open editor tab. This first stage connects the two and lets the agent look: what the user sees and a picture of it. Changing the project comes in stage 2. - An agent token per account opens only /mcp, no API route. The dialog "Agent tokens..." in the account menu creates it once, shows the ready "claude mcp add" line and revokes it. The server keeps only its sha256 (api/lib/AgentTokens.php). - Wrong tokens count into the sign-in brake from #427. Changing the password, renaming or deleting an account ends its tokens. - The relay (tools/mcp/) is a small Node service without packages, WebSocket server included. It runs in the container feuerwerk-mcp because PHP-FPM cannot hold connections. Whether a token or a tab ticket is valid is still decided by the PHP API, through two routes that answer only with the relay key. - The tab joins over WebSocket on the same origin with the normal session, an Origin check and a one-time ticket. It connects only after "Agent" is switched on in the account menu. One tab per account is active, a new one takes over. Guests get nothing. - Tools: get_view (project, tab, document, selection with names, camera, play state, open dialog) and screenshot. The resource feuerwerk://view can be subscribed. Claude Code does not act on that reliably, so a changed view is also noted on the next answer. - A badge next to the account menu shows the connected agents. It counts only agents with an open stream or a request in the last two minutes. - CI starts the relay in front of php -S (tools/ci/serve.sh), and e2e 71 runs the whole path in the browser. Found on the live test: the badge showed four agents because old sessions of revoked tokens were still counted. Tests: - Gate on atx: 316 files, 4888 tests - relay protocol test with fake API and fake tabs: 23/23 - get_view, tools, socketUrl: 16/16 - PHP agents 29, rights 103 checks - e2e 71 on atx 10/10, against the live site 10/10, no CSP violation - real Claude Code against the live site: get_view answered - ZAP against /mcp: nothing at High or Medium - every new check fails when the guard it covers is taken out (6) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
925a39a276 |
Let guests try a demo project without an account (#428)
The editor was only open to somebody with an account. A guest now picks
a demo project on the sign-in screen and gets the whole editor on a copy
held in the tab. The server writes nothing for a guest, so there is no
guest session, no guest folder and no quota to keep.
- Demo copies are frozen copies like a release, in data/demos/. An
administrator makes, updates and removes one with "Demo copy..." in
the project's main menu. GET demos and GET demos/{token} need no
session. An empty project is just a demo that is empty.
- Demo material: an administrator offers an uploaded file with "Offer
as demo" in the pickers (PATCH {demo}, data/demo-files.json). Every
list carries it with demo: true, for accounts and guests alike.
Signed out, a list holds the demo material alone, without the owner.
- An offered file is only an administrator's to delete. A file a demo
copy names counts as used, like one a release names.
- The guest's project lives in MemoryProjectStore. Every store that
would write refuses in the tab before a request is made, and the
autosave is not attached at all.
- The main menu drops projects, snapshots and release for a guest. The
pickers drop Upload and Delete, a dropped or pasted file is refused,
and a recording is saved as a download.
- The toolbar says the work is not saved. The Account menu holds
"Sign in..." instead of Logout and the password.
- A guest's demo never goes into the address: a reload brings the
sign-in screen back, not the copy. Found by e2e 70.
- GET plugins needs no session any more, since a demo may switch one
on. It only reads the server's own modules.
Tests:
- e2e 70 (atx): ALL PASSED, 13 checks. Counter-test with the guest
autosave sent to the server: 70f and 70g red on the PUT.
- tests/core/guest.test.ts: 9 tests. PHP: demos 20, files 46,
rights 80 checks (the matrix asks every new route).
- Gate (atx): CHECK EXIT 0, 314 files, 4852 tests.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
||
|
adminandClaude Opus 5.5
|
cfd6d7a8df |
Lock the server down before the editor goes public (#427)
The editor is about to be opened to strangers: a post on Reddit and a guest access (#428). nginx served the whole workspace and ran every PHP file in it, the API let anybody guess passwords as fast as they liked, and one account could fill the disk. - nginx (shared web stack, outside this repo, backup in /root/config-backup-20261015/) hands out only the two pages, dist/, public/brand/, the uploads by kind and ending, and api/index.php. Everything else under /feuerwerk/ is a 404. - PHP behind it has display_errors off and no X-Powered-By. Plain HTTP forwards to HTTPS, the API refuses it. - Headers: a Content-Security-Policy without inline code or eval, nosniff, frame-ancestors, HSTS, Referrer-, Permissions-, COOP, CORP. - vite.config.ts writes the same policy into the built pages, so CI runs the whole suite under it. Pixi gets by without eval through pixi.js/unsafe-eval. The page CSS left its inline style for a file. - The harness fails a module for anything the policy refuses, from Chrome's log and from the page's securitypolicyviolation event. - LoginThrottle: five wrong passwords from one address within 15 minutes, and it gets 429 until 15 minutes after the last of them. The address is the last X-Forwarded-For entry, the Brandmauer's own. - Quota: 1 GB per account for uploads, projects and snapshots. Only a write that grows the account is refused (413). Admins have no limit. - A write whose Sec-Fetch-Site says another site is refused (403). - The session cookie carries Secure, nginx passes HTTPS on. - showFailure puts the message in as text, not as HTML. - Every PHP script outside api/ stops unless it runs on the command line. - The video worker sees the checkout read-only and can only write data/videos. No capabilities, 2 GB and 256 processes at most. - tests/php/rights.test.php runs the API under php -S on a scratch directory and asks every route as signed out, owner, stranger and admin. A route named in api/index.php but missing in the matrix fails it. limits.test.php covers the brake and the quota. Findings: - tmp/ was readable from outside, with 79 copies of project files. Closed on 15.10. at 20:02 with chmod 750. The log back to 11.10. shows only our own requests. - Port 8080 is bound on 0.0.0.0, so the LAN could reach nginx past the Brandmauer. Left as it is: the web stack is not this project's. - npm audit finds nothing in what is shipped. Dev tools (vitest, brace-expansion, source-map-js) have advisories, not touched here. Tests: check 313 files / 4843 tests. PHP limits 21, rights 60, accounts 56, snapshots 15, releases 20, files 46 (videos 50/51, no ffmpeg in the web container). e2e on stein2, atx was off: 34+38+41 197/198, the one red check (a click in 41) passes with 41 alone, 27/27. 67 misses frames at 1080p on stein2 with and without this. Path list from outside: 10 served, 26 refused. ZAP baseline against the live site: nothing High or Medium under /feuerwerk/. Counter-checks: rights.test.php without the cross-site guard, the quota, the brake, the admin check of users, or with an undocumented route fails each time. A policy without wasm-unsafe-eval fails e2e 41 with the harness's own CSP line. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
6097307b6d |
Move repo, CI and the video worker over to the name Feuerwerk (#422)
Stage 3 of the rename. The repo now lives at meleki/feuerwerk, the runners carry feuerwerk labels and the WSL distro on melektaus-atx is called feuerwerk-ci. The code has to follow, or CI finds no runner and the suite no longer picks the GPU path. - ci.yml: workflow name feuerwerk, jobs on feuerwerk-ci (gate, stein2) and feuerwerk-wsl (suite, melektaus-atx). - GAME_MAKER_* variables are now FEUERWERK_*. This includes FEUERWERK_BROWSER=wsl, which the runner config on melektaus-atx sets. - The video worker runs as container feuerwerk-video with the image feuerwerk-video and works on /srv/feuerwerk. - e2e.sh starts the unit feuerwerk-e2e, the CI image keeps its paths under the new name. - README and repo links point to git.nockal.com/meleki/feuerwerk. The old URLs still redirect with 301. Both runners keep their old labels until the first CI run with this ci.yml is through. Then they go, together with the image tag game-maker-ci:latest and GAME_MAKER_BROWSER in the atx config. Tests: - Gate on melektaus-atx through the renamed distro: CHECK EXIT 0, 307 files, 4756 tests. - e2e 60 on melektaus-atx: ALL PASSED, 20 checks. Renderer is the RTX 4070 through D3D12, so FEUERWERK_BROWSER reaches the GPU path. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
d48f174caf |
Play videos on sprites, converted on the server with clips (#402)
A video becomes an asset like a sound: uploaded once, converted to one format by a worker of its own, cut into clips in its tab, and worn by a sprite through its Texture row. Only Chrome is a target, without alpha. - Server: POST ?r=videos queues the upload under the hash of its bytes, so the MP4's name is known at once. The worker (VideoWorker over VideoFormat) writes H.264/AAC, yuv420p, constant frame rate, a keyframe every 15 frames, at most 1920x1080, with a poster and the details beside it. A failed job says why and is tried again on the next upload. - An upload may be up to 2 GB. It is streamed from the request to disk and hashed there, so php-fpm holds none of it in memory. The other routes keep their 256 MB. - The worker runs in the container game-maker-video on stein2, a compose project of its own (tools/video/), and in CI beside the PHP server. The CI image has ffmpeg for it. - Model: asset type video with size, length, frame rate and sound, and clips like a sound's. The clip helpers and commands serve both. - A sprite's Texture row offers videos and clips after the pictures. resolvePicture names a playing video by a key, and the texture library hands it to a VideoLibrary with one player per key. - The editor shows a placed video as a quiet preview. Play starts it over at its clip's start and lets it be heard through the master bus, a pause holds it, Esc makes it quiet again. - Video tab: the picture, a ruler, the clip strip and a track on the Sound tab's time axis. A drag spans a clip, a handle moves an edge, Play plays the picked clip. - The picker shows a video on its way with its progress and adopts an upload from it once it is ready. Dropping a video file does the same. Decided on my own: - Two sprites wearing the same clip share one player and show the same picture. Control per instance belongs to the follow-up with nodes. - A video plays in real time, the game's time scale does not slow it. - The container may use 3 of the 4 cores of stein2. Findings: - The first real upload (279 MB) got 413: the API's own ceiling of 256 MB for every request, not nginx or the Brandmauer. - Chrome has no picture of a video that never played, and a Pixi VideoSource drawn before the first frame stays empty for good. The texture is made in the first frame callback (openVideo). - A player nobody asks for in 2 s is let go. The inspector's PREVIEW asked only on an edit, so with the sprite not on screen its texture was destroyed under it and the batcher threw on every frame. The preview now redraws every frame while it shows a video, like a Time shader. A let-go texture keeps an empty source instead of none. - VideoSource.load() reloads the element and cuts off an earlier play(). Playing starts after loadeddata. Tests: - PHP videos.test.php: 68 checks, with the real ffmpeg - 300 MB upload through Brandmauer and nginx: 201 in 21 s, php-fpm at 13.7 MB RSS at most - Unit videos.test.ts: 20 tests - e2e 66 (stein2, atx was off): ALL PASSED, 18 checks; e2e 04 with the new asset count green - Gate (atx): 294 files, 4520 tests. serialize.test.ts used video as its example of an unknown type and now uses a made-up one. - Counter-checks: a keyframe gap of 30 fails the PHP check, a key that keeps the video's end fails 2 unit tests, a clip without its end fails "plays only its stretch" in e2e 66 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
6b49a0bea7 |
Give each account its own folder of uploaded files to delete (#344)
Uploaded images, fonts and sounds belonged to nobody. Every account saw every file in the pickers, and only an administrator could delete one, through the API alone. Now every account uploads into a folder of its own, sees its own files and nobody's, and can delete a file of its own that no project uses. - Files are stored as data/<kind>/<folder>/<sha256>.<ext>, and the folder is part of asset.file. A name without a folder belongs to nobody. - The folder is a random 12-character name kept in users.json and given with the first upload. Renaming an account moves no file, and a released game's URLs carry no account name. - The list shows the caller's own files and nobody's. An administrator sees all, each with owner and deletable. - DELETE takes the name with its folder. It answers 403 unless the file is the caller's or the caller is an administrator, and 409 while a project, snapshot or release of any account names it (FileUsage). The 409 names the caller's own projects only. - The pickers have a Delete button. The first press asks, the second deletes, and a refusal shows the server's reason. - A deleted account's folder stays and only administrators see it. - tools/migrate-file-owners.php moved every used file into the folder of the account using it (a copy each for two accounts) and rewrote projects, snapshots and releases. A project's revision goes up, so an open editor gets a conflict instead of saving old names back. - Font families are named after the hash alone, so a folder never ends up in a CSS family name. Run on stein2 on 2026-10-07, backup in /root/config-backup-20261007/data-before-344: 25 files used, 27 copies, 8 projects, 53 snapshots and 1 release rewritten, 30 unused files stay nobody's. Afterwards all 600 references resolve to a file in their owner's folder. Found while building: the Delete button cut its new label to "Del…" because a button does not measure itself again on a new text. The picker now lays its button row out again. Tests: - PHP files.test.php (new): 46 checks, counter-check 3 red - Unit core + editor: 220 files, 3403 tests - e2e 34 (melektaus-atx): ALL PASSED - 67 checks (2 new) - Gate (melektaus-atx): CHECK EXIT 0, 256 files, 3977 tests Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
42647d309b |
Run the gate on stein2 and the suite on the WSL runner (#283)
The gate needs no GPU, so it goes back to the runner on stein2 (`game-maker-ci`, same image). The suite stays on `game-maker-wsl` on melektaus-atx. The two jobs no longer share the four cores of one runner while they overlap. - The change of `runs-on` for the gate is the user's own edit. - The header comment now says which job runs where. GM_TEST_WORKERS stays at two, for the day both jobs share one runner again. Tests: - Both runners were online in Gitea before the push (`stein2-docker`, `melektaus-atx-wsl`). - Prettier passes on the file. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
857e23c69b |
Run CI on a WSL2 runner with the RTX 4070 (#283)
stein2 should stop carrying the full e2e run. The new runner lives in the WSL2 distro game-maker-ci on melektaus-atx and uses the same image and fonts. Its GPU sits behind /dev/dxg and Mesa's d3d12 driver, which headless chromium cannot reach, so the harness gets a second way to start the browser. - GAME_MAKER_BROWSER=wsl starts an invisible Xvfb and an X11 chromium on ANGLE/GL. Without the variable, stein2 keeps the headless Vulkan launch exactly as before. - The harness logs the renderer and stops on software rendering. Under WSL a missing d3d12 driver falls back to llvmpipe, which still offers WebGL 2 and would pass unnoticed. - The CI image gains xvfb. The font set is unchanged: fc-list gives the same 149 fonts with the same hash. - ci.yml runs both jobs on the new label game-maker-wsl. The stein2 runner keeps game-maker-ci, so moving back is one line per job. Findings: - Wayland on weston reached the GPU too, but drew at about 40 fps and let the editor clock stall in long runs (GM_STEPS=1-17: 2 failed). - Vulkan through a self-built dzn caps ANGLE at ES 2.0 because VK_EXT_provoking_vertex is missing, so chromium drops the GPU. - The flaky check in e2e 41 (#282) fails more often under WSL: two of three solo runs were red. Tests: - ./check.sh: 195 files, 2823 tests passed - WSL container, GM_STEPS=1-17: ALL PASSED, 553 checks - WSL container, full run: 1589 of 1590, the one FAIL is #282 - WSL container, smokes: 24 / 19 / 21 checks - stein2 headless, e2e 39+40: ALL PASSED, 100 checks Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
4194f3e150 |
Share a link that plays the game without the editor (#266)
A project could only be played inside the editor, behind the login.
To show a game to someone else there was no way short of an account.
A release freezes the saved project under a random token and serves
it on a page that holds nothing but the game view.
- Menu -> Release... opens a dialog with the state of the release,
the link with Copy link and Open, Publish or Update release, and
Take down behind a confirmation.
- Publish saves first and refuses while anything is unsaved, like
snapshots. Updating keeps the link. Take down kills it, and a new
Publish hands out a new token.
- The link is play.html?game=<32 hex>. The page needs no session,
paints Theme.colors.background, zooms with Auto and waits for
"Click to play" so that sound and keyboard focus are allowed.
- The game always starts in the start scene. Persistent variables
are stored per link in the browser. Plugins are not loaded, and the
dialog warns about them and about a project without a scene.
- GET ?r=releases/{token} is the only route besides auth/* that works
without a login. The public answer carries the token as project.id,
so the project id stays hidden. Release files are written with 0640
and nginx does not serve them directly (measured 403).
- Starting, the frame and Load scene moved into runtime/GameRun.ts.
Play in the editor and the play page run through the same code.
- WorldViewport moved to src/viewport/ and no longer needs the
EditorState. Pointer forwarding (gamePointer.ts) and setting up
textures, fonts and sounds (projectFiles.ts) exist only once.
- Vite builds two pages. The play page loads 4 kB of its own code
plus the shared chunk (Pixi, runtime, drawing, 653 kB) and Rapier,
without the editor.
Found on the way, not fixed here: data/projects/<id>.json answers 200
without a login because the files are 0644.
Tests:
- ./check.sh: 194 files, 2808 tests, exit 0.
- tests/php/releases.test.php: 20 checks, also added to CI.
- New e2e module 60 with 10 checks, played in a browser without a
session. Module 46 counts the menu entries and was updated.
- Full e2e run: 1572 checks passed.
- Counter-checks: 3 unit, 2 PHP and 3 browser mutations all fail.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
||
|
adminandClaude Opus 5.5
|
94749b4f3e |
Let a project keep named snapshots and restore them (#231)
Autosave only keeps the latest revision. Before a risky change there
was no way to put a known good state aside and go back to it later.
The main menu now has "Snapshots..." to create, restore and delete
named copies of the open project.
- Server: SnapshotStore.php keeps each snapshot as a copy of the
project JSON under data/snapshots/<project>/ (git-ignored). Four
routes under projects/{id}/snapshots use the same owner check as the
project, so an admin also sees foreign snapshots.
- Create writes the autosave first. With unsaved edits (play with open
edits, failed save, conflict) no snapshot is made and a message says
why. The rule is the pure function in src/editor/snapshots.ts.
- An empty name falls back to the local date and time.
- Restore asks first, then keeps the current state as "Before
restoring <name>". It writes the snapshot back as a new revision and
reloads the project, so the undo list is empty. A restore can itself
be undone by restoring that snapshot.
- Deleting a project deletes its snapshots too. Images, sounds, fonts
and pixel art are content-addressed and never deleted, so a snapshot
can always reach them.
- Client: SnapshotStore and HttpSnapshotStore in src/core/store, the
SnapshotsDialog, and the flows in ProjectSession. A stale revision
(409) becomes StaleRevisionError.
- CI runs the new PHP test in the gate job next to accounts.
Not covered by e2e: the lock during play and the conflict case. Both
are unit tests only.
Tests:
- check.sh: 164 files, 2401 tests, CHECK EXIT 0.
- tests/php/snapshots.test.php: 14 checks, all passed.
- tests/editor/snapshots.test.ts: 5 tests for the time format, the
blocker rule, the routes and 409.
- New e2e module 50 on a throwaway project: create, rename, restore,
undo the restore, delete, then delete the project and get 404 on
its snapshots. 46b expects the new menu entry. e2e 46,50: 20 checks
passed. The check sum goes from 1385 to 1393 (computed).
- Counter-checks, originals restored and compared with cmp: blocker
always allowed fails 1 unit test. No sorting fails 2 PHP checks.
Restore without keeping the old state fails 50c and aborts 50d.
- The first sort test with 2 snapshots passed by chance half the time
because of random ids. It now uses 8.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
||
|
adminandClaude Opus 5.5
|
ceaa44e5b8 |
Add roles and a user management tab for admins (#180, #179)
Until now every account could do the same, and managing accounts meant running make-user.php on the server. Accounts now carry a role, and admins get a tab in the editor to create, rename, re-role, reset and delete accounts. Both issues share the same server files, so they land in one commit. #180 Roles: - Every account is admin or user. A missing or unknown role counts as user, on the server and in the client. - auth/session and auth/login return {user, role}. - Auth::requireAdmin() answers everyone else with 403. - make-user.php takes the role as a third argument. Without it an existing account keeps its role, so a password reset from the CLI never demotes an admin. - Admins see and save other users' projects like their own. Saving keeps the original owner. #179 User management: - The account menu gets "User management" (admins only) and "Plugins" (everyone). Both open the same closable tab, accounts on top and plugins below. The fixed Plugins tab is gone. - Per account: Rename, Role, Password (no old password needed) and Delete with a confirmation. "New user..." creates one. - Each account lists its projects with an Open button. - Delete is refused while the account still owns projects. - Renaming moves the projects along and ends every session of the account. transfer takes the same lock as save. - On their own row an admin can only set the password. The last admin cannot be demoted or deleted. The server enforces both. - The account rules live in AccountRules.php and accountRules.ts as pure functions, so they are testable without a server. Findings: - At 1280 px the Delete button reached past the column edge and was not clickable. The name column is flexible now, and module 49 checks the layout. - The server lets an admin delete other users' projects, but the editor has no button for it yet. Left open in the issue. Tests: - ./check.sh: 148 files, 2214 tests, exit 0 (was 147 / 2204). - New tests/php/accounts.test.php, 51 checks, plain PHP. CI runs it as its own step in the gate job, the host has no PHP. - New e2e module 49 with 18 checks. CI creates the e2e-admin account in serve.sh. Modules 14, 38, 46 and the scaffold follow the new tab. - e2e 1-16,37,38,45,46,49: all passed, 581 checks. - Counter-checks: owner on save, admin view, default role and self delete in PHP fail 7, 5, 5 and 1 checks. Unchecked requireAdmin, a wider column and the caller as owner each fail exactly their e2e check. - Check sum 1314 -> 1333 (+1 in 38, +18 in 49), computed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5
|
15bdfdf07a |
A gate that stands in front of the suite delays the news it carries, so it runs beside it (#94)
The workflow ran everything in one job: check.sh, then build, then the e2e suite and the smokes. The runner has two job slots and four cores, so the second slot lay idle for the whole run, and a lint error was reported only after the full suite had passed in front of it - six minutes instead of one. Now `gate` and `suite` are two jobs without `needs:`. They start together, both have to pass. The price is contention: while the gate runs, vitest and chromium share four cores, and the suite measures text widths and frame counts, which a loaded machine reports differently. The gate therefore caps vitest at two workers via GM_TEST_WORKERS, so two cores stay free for the browser over the whole overlap. Locally the variable is unset and nothing changes. Measured before the push: the cap costs vitest 4.4s (16.6s -> 21.0s), check.sh exits 0 with it, build ok. Whether the split is worth it is a question for the runner - the expected net gain is about 50s, and the acceptance is a measurement over at least two runs. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
rootandClaude Opus 5
|
4f0908df77 |
Run the gate, the editor suite and the three games on the Gitea runner
game-maker / suite (push) Successful in 7m31s
The job image (tools/ci/Dockerfile, label game-maker-ci) carries chromium with the Vulkan drivers, php-cli and the host's fonts; the render device itself comes from the runner's container.options, so a job container gets the Intel GPU instead of llvmpipe. tools/ci/serve.sh puts the checkout behind /game-maker/ with php -S and writes the project shells the seeders need; every tool now takes its server from GAME_MAKER_BASE_URL rather than from a project id repeated in shell scripts. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |