Files
magpie/internal
yetone c431225a08 gateway: a browser that reaches loopback under a web site's name (DNS rebinding) is refused unless it sends an enabled gateway key
What changed
- lanGuard refuses (403) a loopback request that carries a browser's
  marks (Origin, Sec-Fetch-Site/Mode, a Mozilla/ user agent) when its Host
  isn't this computer's: an IP, localhost, *.localhost, a container
  runtime's name for the host (host.docker.internal and kin),
  MAGPIE_PUBLIC_URL's or MAGPIE_ADDR's host, or the machine's own name.
  An enabled gateway key still lets it in.
- Why: a page whose DNS answers 127.0.0.1 after it loads is same-origin
  with the gateway. Its GET sends no Origin and its POST one equal to the
  Host, so corsGuard's foreignPage let it by, and on loopback no key is
  asked. The nightly security review reproduced it on origin/main: Host
  and Origin rebind.attacker.example:3425 from 127.0.0.1 got 200 on
  GET /v1/models, POST /v1/chat/completions (spends subscriptions) and
  GET /v1/magpie/quotas (account names, balances). The subsystem doc
  listed it as a known gap.
- Agents send none of a browser's marks and pass under any name; browsers
  and webviews calling localhost or an IP are unchanged.
- Tests: TestCORSKeyThroughTheServer and the Omarchy test now set a real
  Host (httptest's default is example.com). TestForeignPagesRefused's
  "own origin behind a port-443 name" row from loopback with no forwarding
  header now sends a key: without one it is the rebound page.
- docs/subsystems/gateway-routing.md describes the check in place of the gap.

Ran
- TestReboundPageRefused: fails on origin/main ("a rebound page's GET:
  200"), passes with the change.
- go test -count=1 ./internal/gateway/ ./internal/gui/ ./internal/access/
  (macOS, go1.27): ok. gofmt -l clean; GOOS=linux and GOOS=windows go vet
  ./internal/gateway/ clean.

Not tried: a real browser against a real rebinding DNS name.
2026-10-10 14:21:23 +08:00
..
settings: keep awake can keep the display on too (#975, Hu9956: magpie's keep-awake held the system up, but the screen still went black and locked, and their agents record the screen to check their work, which a locked screen stops) — Settings › General's "Keep awake while agents work" row now has three segments, Off / On / Screen on too, saved as keepAwake and the new keepAwakeDisplay (default off, so On and every existing settings.json behave as before). awake.Keep takes a Level (Off, System, Display) in place of a bool; a level changed while held releases and takes the hold anew. On a Mac the display level runs caffeinate -i -d -w <pid> (pmset -g assertions on a caffeinate of the same flags shows PreventUserIdleDisplaySleep beside PreventUserIdleSystemSleep); on Windows SetThreadExecutionState gets ES_DISPLAY_REQUIRED too; on Linux logind's idle inhibitor (systemd-inhibit --what=idle:sleep, unchanged) is all there is, and whether the screen stays on is up to the desktop. keepAwakeDisplay is this computer's own (KeepOwn) and goes in prefsKeep, so another setting's save keeps it. The row's line says the display stays on when it's chosen; it keeps its English in data-en, so a language change re-translates it. zh/ja/de strings added. No TUI or CLI has keep-awake, so none changed. TestKeepDisplay (taken once, taken anew with the display, not again at the same level, back to System, released at Off), TestCaffeinateArgs, TestExecutionState (GOOS=windows test build) and TestKeepAwakeKeptAndOwn cover it; go test ./internal/awake ./internal/settings ./internal/gui ./internal/gateway pass, -race -count=5 on awake. keep-awake.test.cjs now runs en/zh/ja/de at 900px and 440px in Chromium and WebKit (display posts keepAwake+keepAwakeDisplay true, the line changes, another setting's save keeps it, segments fit the row, On and Off clear it, no click moves the page): 32/32 twice; gui-ja, gui-de, i18n-keys and click-scroll pass. go vet (darwin, windows, linux), darwin, linux (nogui) and windows builds pass. Not tried with an agent recording the screen; the reporter is asked to confirm.
2026-10-06 15:33:28 +08:00
windows: an account list read while another magpie writes it is waited for, not taken for no accounts (Packing1 on Discord, Windows 11: 早上更新重启后之前登录的Antigravity订阅供应商和账号都没了 … 整行都没了,供应商里面antigravity直接消失了) — on Windows a file being renamed over (every write magpie makes: a temp file renamed onto it) is, for a moment of up to ~2ms, "The system cannot find the file specified" to anyone else looking at it, with MoveFileEx and with POSIX-semantics renames alike. filememo.Read gave that back as an error, readLogins as no accounts, and whatever change came next (any account's read-modify-write) wrote the empty list back: every built-in account gone, its subscription with it. Measured on Windows 11 (10.0.0.132): one process replacing a 200 KB file, another stat-ing and reading it, saw the file "not there" in a few windows each 20 s. Now steady.Stat and steady.ReadThere wait out a moment of a file not there (15 tries a millisecond apart, Windows only), and filememo looks through them, remembering a file found not there so one that really isn't costs the wait once. A magpie started by an update now waits for the one it replaces before its start-up moves write any file (AwaitPredecessor at the top of run, not only in the GUI). TestLoginsKeptWhileAnotherMagpieWrites (Windows: two writers replacing logins.json, readLogins never without the Antigravity account) failed 1 in 3 and 2 in 8 runs on the old code ("read as no accounts 8 times"), passed 11 of 11 on this; filememo's TestReadWhileReplaced the same (1180 misses on the old code); TestStatWaitsOutAReplace everywhere. A real update in a sandbox home on the Windows box, magpie-cli 0.1.880 → 0.1.942 from the release, kept the account and logins.json byte for byte. go vet (darwin, and windows nogui), builds (darwin, GOOS=linux nogui, GOOS=windows) and go test -tags nogui -count=1 ./... under a temp HOME pass.
2026-10-05 10:57:45 +08:00
…
plugin, gateway, gui: a plugin can be gateway middleware — onRequest, onEvent and onResponse run in magpie's gateway on what agents send and get back (moejs) — a plugin could only be a subscription provider, run in the Bun host, a process away from the gateway. A plugin may now declare middleware (package.json magpie.middleware, or a *.middleware.js / .mjs file), run in-process by moejs (pure-Go JS engine): compiled once, one runtime per request from a pool, so every hook of a request shares ctx.state. onRequest(body, ctx) sees the agent's body in its own API before routing; what it returns is the body the next middleware and the gateway get, undefined leaves the bytes as they were, ctx.reject(status, msg) answers with an error in the agent's API's shape. onEvent(ev, ctx) sees each SSE event's JSON (export const events = [...] filters by name; an object changes it, null drops it, undefined passes the bytes through untouched; data: [DONE] is no JSON and passes by). onResponse(body, ctx) sees a whole JSON reply, with ctx.status. ctx has protocol, model, stream, path, agent, the plugin's options and state. A hook that throws, returns a non-object or runs past its limit (250ms request/response, 50ms an event, 1s load; interrupted) leaves the request or reply as it was and is counted and logged (60 lines a minute); a compressed reply passes as-is. Files are reloaded within a second of a change. Only agents' requests run through it (serveAgent at the gateway's agent entry points and the Codex backend), never magpie's own titles, classifier or stand-ins; it sees bodies before redaction. A middleware-only package is skipped by the Bun host. The Plugins page marks a middleware with a chip, its hooks, calls and µs per call (JSON in and out counted), failures in red with the last error, and a load error. en/zh/ja/de; docs/reference.md and usemagpie.ai/docs/plugins#middleware (en/zh/ja) with an alias example. Tested with a real DeepSeek account through magpie web --gateway in a sandbox HOME: an alias middleware served "fast" and named it in message_start, a reject came back as 403 permission_error, a Chat stream uppercased; real Claude Code (sandbox CLAUDE_CONFIG_DIR) did Read, Bash and Glob tool turns through three middlewares with 0 failures: ≈1–1.6µs an event, 0.2–0.7ms an onRequest that parses and rewrites Claude Code's ≈80KB body. Real use found two bugs fixed here: [DONE] parsed as JSON (SyntaxError on every Chat stream) and stats leaving out the JSON time. internal/middleware tests (chain, undefined keeps bytes, reject, fail-open on throw/spin/string with the runtime reused, async hook, load errors, off and Bun skip, reload, split/CRLF/filtered/dropped stream events, [DONE] untouched, whole reply and gzip, 64 in parallel; BenchmarkEvent ≈0.9µs, 11 allocs) and TestMiddlewareThroughGateway (alias routed, events by protocol, redaction still upstream, reject in Anthropic's shape; fails without the change) are new; plugin-middleware.test.cjs (Chromium and WebKit, en and zh: chip, hooks and stats, failures with last error, broken and off, no left border) fails without it. The other plugin GUI tests and click-scroll pass in WebKit. go vet, go build (darwin, GOOS=linux nogui, GOOS=windows) and go test -tags nogui ./... under a temp HOME pass (internal/provider failed once under load and passes alone).
2026-10-05 19:14:31 +08:00
windows: an account list read while another magpie writes it is waited for, not taken for no accounts (Packing1 on Discord, Windows 11: 早上更新重启后之前登录的Antigravity订阅供应商和账号都没了 … 整行都没了,供应商里面antigravity直接消失了) — on Windows a file being renamed over (every write magpie makes: a temp file renamed onto it) is, for a moment of up to ~2ms, "The system cannot find the file specified" to anyone else looking at it, with MoveFileEx and with POSIX-semantics renames alike. filememo.Read gave that back as an error, readLogins as no accounts, and whatever change came next (any account's read-modify-write) wrote the empty list back: every built-in account gone, its subscription with it. Measured on Windows 11 (10.0.0.132): one process replacing a 200 KB file, another stat-ing and reading it, saw the file "not there" in a few windows each 20 s. Now steady.Stat and steady.ReadThere wait out a moment of a file not there (15 tries a millisecond apart, Windows only), and filememo looks through them, remembering a file found not there so one that really isn't costs the wait once. A magpie started by an update now waits for the one it replaces before its start-up moves write any file (AwaitPredecessor at the top of run, not only in the GUI). TestLoginsKeptWhileAnotherMagpieWrites (Windows: two writers replacing logins.json, readLogins never without the Antigravity account) failed 1 in 3 and 2 in 8 runs on the old code ("read as no accounts 8 times"), passed 11 of 11 on this; filememo's TestReadWhileReplaced the same (1180 misses on the old code); TestStatWaitsOutAReplace everywhere. A real update in a sandbox home on the Windows box, magpie-cli 0.1.880 → 0.1.942 from the release, kept the account and logins.json byte for byte. go vet (darwin, and windows nogui), builds (darwin, GOOS=linux nogui, GOOS=windows) and go test -tags nogui -count=1 ./... under a temp HOME pass.
2026-10-05 10:57:45 +08:00
upstream: the vendors' own status pages say when their API is down (#971, emo172: an Anthropic or OpenAI outage looked like a sign-in or quota problem) — a provider that calls Anthropic's, OpenAI's or Moonshot's API (by its base URL's host: api.anthropic.com, which the Claude subscription calls too; api.openai.com and chatgpt.com, Codex's; api.moonshot.cn/.ai) carries "API outage" (red), "API degraded" or "API maintenance" beside its name on its Providers row and its Usage card while the vendor's page says so; its title names the parts and the open incidents, and a click opens the incident (or the page), not the row. internal/upstream reads each page's Statuspage-style summary.json (status.claude.com; status.openai.com, incident.io's; status.moonshot.cn), checked by hand on 2026-10-06, and only the API's own parts (Claude API and Claude Code, not claude.ai; Chat Completions, Responses, Codex, Codex CLI, not "Codex in ChatGPT Desktop", which was degraded that day); an incident naming no parts (OpenAI's) counts only while those parts are affected. Pages are read only when the Providers or Usage page asks (GET /api/upstream), at most once in five minutes, never on a request's way; a page that fails, isn't JSON or no longer lists the API's parts is unknown and marks nothing. A relay serving the same models is never marked. DeepSeek's page (Flashduty) has nothing readable and is only known for its link; OpenRouter, xAI, Mistral and z.ai have no page read. The gateway's failure messages, TUI, CLI and tray aren't changed. zh/ja/de added. upstream_test (fake pages from the real summaries: API down, claude.ai down isn't the API, Claude Code degraded, OpenAI's app isn't the API, 503/HTML/changed page unknown, read once in Fresh and Of never waits, hosts; -race -count=5) and gui's TestUpstreamRoute pass; upstream-status (Anthropic down: its provider and the Claude subscription marked, relay/OpenAI/unread Moonshot not, click opens the incident not the row, all well marks nothing; 900 and 440px, the badge whole, the name giving way) passes in Chromium and WebKit, en and zh, and fails 8/8 without the app.js change. i18n-keys, gui-ja, gui-de, click-scroll, account-return, account-models, usage-keys-fold, deprecated-subscriptions and add-sheet pass; go vet, go test ./internal/gui/, linux/windows/nogui builds pass.
2026-10-06 12:36:05 +08:00
…