mobygate 0.8.1 → 0.8.3

package/CHANGELOG.md

@@ -4,6 +4,148 @@ All notable changes to mobygate are documented here. Format loosely follows
 [Keep a Changelog](https://keepachangelog.com/en/1.1.0/); version numbers are
 [Semantic Versioning](https://semver.org/).
 
+## [0.8.3] — 2026-04-28
+
+OpenClaw .26 compatibility, sonnet 1M context, and observability for
+silent model swaps. Found by upgrading mobygate to v0.8.1 on Windows
+during heavy multi-agent testing.
+
+### Fixed
+
+- **OpenClaw connector wrote deprecated `models.main` / `models.default`
+  keys**, which OpenClaw `2026.4.26+` outright rejects with `Unrecognized
+  keys: "main", "default"`. The gateway refused to start until
+  `openclaw doctor --fix` cleaned them up.
+
+  v0.8.3 connector now writes the modern path:
+  - `agents.defaults.model.primary` (the active default)
+  - `agents.defaults.models` (registered model id map)
+
+  Plus on every `plan()` it removes the deprecated top-level keys if a
+  previous mobygate version (v0.8.0–v0.8.2) wrote them. Re-running
+  `mobygate connect openclaw` after upgrading will repair any broken
+  config in place.
+
+  Why this didn't bite Mac users: hand-rolled configs that never went
+  through `mobygate connect`'s apply path were never written by the
+  bad code. It surfaced on Windows when `mobygate init` auto-wired
+  detected clients during a fresh upgrade.
+
+- **Sonnet 4.6 silently capped at 200k context.** v0.8.2 routed
+  `claude-sonnet-4-6` through directly but didn't append the `[1m]`
+  suffix the way opus 4.7 does, so the SDK ran sonnet at the default
+  200k window. Confirmed via `modelUsage.contextWindow=200000` in the
+  diagnostic logging.
+
+  Fix: `claude-sonnet-4-6` now maps to `claude-sonnet-4-6[1m]`. Verified
+  next-turn `modelUsage` shows `contextWindow: 1000000`. Matches the
+  capability we already advertise in `/v1/models`.
+
+  Side note: `claude-sonnet-4-6-200k` added as an explicit alias for
+  callers that want the cheaper 200k variant.
+
+### Added
+
+- **`[model-billed]` diagnostic log line.** Every successful (non-tool-
+  use-aborted) request now logs the SDK's `modelUsage` map showing
+  what model Anthropic actually billed against. Output looks like:
+
+  ```
+  [model-billed] requested=claude-sonnet-4-6[1m]
+  modelUsage={"claude-sonnet-4-6[1m]":{"inputTokens":3,"costUSD":0.029,
+              "contextWindow":1000000,...}}
+  ```
+
+  This was originally added as a temporary diagnostic to chase whether
+  Anthropic was silently swapping sonnet → opus. The data confirmed
+  no swap is happening — but the line is too useful to remove. Surfaces
+  cost-per-turn, actual context window, and any future silent model
+  changes the SDK might introduce. Low log volume (one line per
+  result message, ~50% of requests since tool_use turns abort
+  before result).
+
+  Bonus finding from this log: the SDK transparently uses **two models
+  per opus turn** — a haiku for fast intermediate work and opus for
+  the final response. Adds ~$0.024 per opus turn beyond what the
+  capture summary shows. Worth knowing for cost analysis.
+
+### Notes
+
+The "claude.ai/settings/usage Sonnet only stuck at 0%" mystery turned
+out to be a Claude Max plan accounting design: the "Sonnet only" bar
+is overflow that only ticks once "All models" is exhausted. On Max
+plans, sonnet usage rolls into the "All models" bar (which was
+climbing as expected). Not a mobygate bug — investigation surfaced
+the v0.8.2 [1m] context omission, which IS a mobygate bug, so net
+positive on the chase.
+
+## [0.8.2] — 2026-04-28
+
+Multi-agent fixes. Found the day after v0.8.1 shipped, while testing
+three OpenClaw bots (Mobius/Lux/Mercury) in parallel on the same
+machine. Both bugs were invisible without the v0.8.1 inspector.
+
+### Fixed
+
+- **Session-key collision when multiple agents share a boilerplate
+  prefix in their system prompt.** v0.7.1's auto-derive hashed the
+  first 500 characters of the system prompt; OpenClaw's "You are a
+  personal assistant running inside OpenClaw…" preamble fills more
+  than that, so the per-agent personality content (loaded later from
+  workspace SOUL.md / IDENTITY.md / etc.) didn't reach the hash. Two
+  separate agents (Lux on sonnet-4-6, Mercury on sonnet-4-6) collided
+  onto the same auto-key when given the same first user message
+  ("@Lux @Mercury Hi"). Same key → same SDK session reuse → cache
+  thrash and potential session-state mixing.
+
+  Bumped `SYSTEM_TRIM` from 500 → 20000 chars. Verified against real
+  captured request bodies that collided in v0.8.1 — they now hash to
+  distinct keys (`auto_b0371e5c…` vs `auto_2b90afd7…`).
+
+  SHA-256 cost on 20kB is ~10-20µs per request, irrelevant in the
+  hot path.
+
+- **Model map silently downgraded `claude-sonnet-4-6` to retired
+  `claude-sonnet-4-5-20250929`.** When the v0.8.0 model map was
+  written, the Claude Agent SDK didn't recognize the un-dated
+  `claude-sonnet-4-6` alias and we worked around it by routing to the
+  most recent dated 4-5. The SDK has since added native 4-6 support,
+  but mobygate kept the workaround in place. Result: clients (OpenClaw
+  Lux/Mercury) configured for sonnet-4-6 were having their requests
+  rewritten to the retired 4-5-20250929 dated id. Anthropic accepted
+  the call but the response wasn't billing into the user's "Sonnet
+  only" quota — it was showing 0% used despite live traffic. Likely
+  Claude was falling back internally to opus or returning a
+  zero-billed degraded response.
+
+  Fix: route `claude-sonnet-4-6` through directly. Also updated
+  `claude-sonnet-4` and the `sonnet` shorthand to point at 4-6
+  (current latest) instead of the retired dated 4-5 entry. Explicit
+  `claude-sonnet-4-5` requests still route to the dated id for
+  backward compatibility.
+
+  Discovery: the inspector showed Lux/Mercury captures all stamped
+  with `model: claude-sonnet-4-6` (correct from the request side) but
+  Anthropic's quota panel reported 0% sonnet usage. The server.log's
+  `model=claude-sonnet-4-6 → claude-sonnet-4-5-20250929` translation
+  line was the smoking gun.
+
+### Notes
+
+The proper long-term fix is for clients to pass an explicit
+`X-Session-Id` header per agent (mobygate has supported this since
+v0.7.1 — it always wins over auto-derive). This bump is a defensive
+measure for clients that don't.
+
+Discovery flow is a nice validation of the v0.8.1 inspector: the
+collision was invisible at the OpenClaw level (each bot's replies
+arrived correctly because OpenClaw maintains its own per-agent SDK
+state) but jumped out as soon as the captures were sorted by session
+key in the inspector — two different model requests with the same
+session-key, with bootstrap text 55kB long but identical first 500
+chars. Without the inspector, this would have surfaced as
+unpredictable cache hit rates and been blamed on Anthropic.
+
 ## [0.8.1] — 2026-04-27
 
 Diagnostic visibility release. Adds a request/response capture system,

package/lib/anthropic.js

@@ -36,17 +36,23 @@ import { v4 as uuidv4 } from 'uuid';
  * against shape variations (the Claude Agent SDK sometimes nests these
  * under `.usage`, sometimes places them flat on the message). Returns
  * a complete usage shape with cache_read / cache_creation fields zeroed
- * out if absent. Used by the 4 mobygate handlers to populate response
- * captures and dashboard cache-hit metrics.
+ * out if absent, plus `modelUsage` (the SDK's per-model usage breakdown,
+ * keyed by the actual model name Anthropic billed against — useful for
+ * spotting silent model fallbacks where the requested model differs
+ * from what Anthropic actually ran).
+ *
+ * Used by the 4 mobygate handlers to populate response captures and
+ * dashboard cache-hit metrics.
  */
 export function extractSdkUsage(message) {
-  if (!message) return { input_tokens: 0, output_tokens: 0, cache_read_input_tokens: 0, cache_creation_input_tokens: 0 };
+  if (!message) return { input_tokens: 0, output_tokens: 0, cache_read_input_tokens: 0, cache_creation_input_tokens: 0, modelUsage: null };
   const u = message.usage || message;
   return {
     input_tokens: u.input_tokens || 0,
     output_tokens: u.output_tokens || 0,
     cache_read_input_tokens: u.cache_read_input_tokens || 0,
     cache_creation_input_tokens: u.cache_creation_input_tokens || 0,
+    modelUsage: message.modelUsage || null,
   };
 }
 

package/lib/connectors/openclaw.js

@@ -169,8 +169,12 @@ export const openclawConnector = {
       configPath: det.configPath,
       mobyProviderExists: !!providers[PROVIDER_NAME_OPENAI],
       mobyNativeProviderExists: !!providers[PROVIDER_NAME_ANTHROPIC],
-      currentMain: det.parsed?.models?.main || null,
-      currentDefault: det.parsed?.models?.default || null,
+      currentPrimary: det.parsed?.agents?.defaults?.model?.primary || null,
+      // Old keys: still surfaced so the user/dashboard can spot if a
+      // pre-v0.8.3 connector run left these around (they're invalid in
+      // OpenClaw .26+). A non-null value here is a "needs reconnect" hint.
+      deprecatedMain: det.parsed?.models?.main || null,
+      deprecatedDefault: det.parsed?.models?.default || null,
       shadowProviders, // pre-v0.8.0 entries pointing at our base URL
     };
   },
@@ -212,14 +216,52 @@ export const openclawConnector = {
           : null;
       if (preferredProvider) {
         const target = `${preferredProvider}/claude-opus-4-7`;
-        after.models.main = target;
-        after.models.default = target;
+
+        // Modern OpenClaw schema (>=2026.4.x): defaults live under
+        // `agents.defaults.model.primary` and `agents.defaults.models`.
+        // The old `models.main` / `models.default` top-level keys were
+        // valid in earlier OpenClaw versions but `.26+` rejects them
+        // ("Unrecognized keys: main, default"). v0.8.0/0.8.1 connector
+        // wrote the old shape and broke OpenClaw `.26` startup. Fixed
+        // in v0.8.3 by switching to the modern path + cleaning up any
+        // old keys it previously wrote.
+        if (!after.agents)                  after.agents = {};
+        if (!after.agents.defaults)         after.agents.defaults = {};
+        if (!after.agents.defaults.model)   after.agents.defaults.model = {};
+        if (!after.agents.defaults.models)  after.agents.defaults.models = {};
+
+        after.agents.defaults.model.primary = target;
+        // Register every model we surface in the provider so OpenClaw
+        // sees them in agents-list. Keep existing entries to preserve
+        // user-registered models (ollama, anthropic-direct, etc).
+        for (const m of MODELS_NATIVE_SURFACE) {
+          const id = `${preferredProvider}/${m.id}`;
+          if (!after.agents.defaults.models[id]) {
+            after.agents.defaults.models[id] = {};
+          }
+        }
+
+        // Cleanup: remove the deprecated top-level keys if a previous
+        // connector run (v0.8.0/0.8.1) wrote them. OpenClaw `.26`
+        // rejects the config outright if these are present.
+        if (after.models?.main !== undefined)    delete after.models.main;
+        if (after.models?.default !== undefined) delete after.models.default;
       }
     }
 
     const summary = diffSummary(
-      { providers: before.models?.providers, main: before.models?.main, default: before.models?.default },
-      { providers: after.models.providers,  main: after.models.main,  default: after.models.default },
+      {
+        providers: before.models?.providers,
+        primary: before.agents?.defaults?.model?.primary,
+        deprecatedMain: before.models?.main,
+        deprecatedDefault: before.models?.default,
+      },
+      {
+        providers: after.models.providers,
+        primary: after.agents?.defaults?.model?.primary,
+        deprecatedMain: after.models?.main,        // should be undefined post-fix
+        deprecatedDefault: after.models?.default,  // should be undefined post-fix
+      },
     );
 
     return {
@@ -263,14 +305,32 @@ export const openclawConnector = {
         if (providers[name]) { delete providers[name]; changed = true; }
       }
     }
-    // If main/default was pointing at us, blank them — let the user
-    // re-pick rather than guess at a replacement.
-    if (isMobyDefaultPointer(after.models?.main)) {
-      after.models.main = null;
+    // Modern path: if the agents.defaults.model.primary points at us,
+    // blank it out — let OpenClaw fall back to whatever default the
+    // user has configured otherwise.
+    const primary = after.agents?.defaults?.model?.primary;
+    if (isMobyDefaultPointer(primary)) {
+      after.agents.defaults.model.primary = null;
+      changed = true;
+    }
+    // Remove our model registrations from agents.defaults.models.
+    if (after.agents?.defaults?.models) {
+      for (const key of Object.keys(after.agents.defaults.models)) {
+        if (isMobyDefaultPointer(key)) {
+          delete after.agents.defaults.models[key];
+          changed = true;
+        }
+      }
+    }
+    // Legacy cleanup: remove deprecated top-level main/default if present
+    // (left over from v0.8.0/0.8.1 connector). OpenClaw .26 rejects them
+    // outright, so removing on disconnect is the right move.
+    if (after.models?.main !== undefined) {
+      delete after.models.main;
       changed = true;
     }
-    if (isMobyDefaultPointer(after.models?.default)) {
-      after.models.default = null;
+    if (after.models?.default !== undefined) {
+      delete after.models.default;
       changed = true;
     }
 

package/lib/session-derive.js

@@ -40,6 +40,12 @@
  *     user message from history mid-conversation, the auto-key changes
  *     and the SDK starts a new session. One turn of double-billing,
  *     then we're back on the new key. Acceptable.
+ *   - **Multi-agent collisions** (fixed in v0.8.2): two agents that
+ *     share boilerplate at the start of their system prompt previously
+ *     collided onto one session key when the trim window only covered
+ *     the boilerplate. SYSTEM_TRIM was raised from 500 to 20000 chars
+ *     to capture the per-agent personality content that follows the
+ *     shared preamble. See note on the constant below for details.
  *
  * Opt-out: `X-Session-Id: none` tells us the client explicitly wants
  * stateless behavior — we return null and the request flows through
@@ -51,7 +57,20 @@
 import { createHash } from 'crypto';
 
 const HASH_LEN = 16;
-const SYSTEM_TRIM = 500;
+// SYSTEM_TRIM was 500 in v0.7.1 — large enough for casual single-agent
+// scenarios (Hermes, single-bot OpenClaw) but caused collisions when
+// multiple agents shared a common boilerplate prefix. Observed in v0.8.1
+// production: Lux + Mercury (two OpenClaw agents) both started their
+// system prompt with the OpenClaw "You are a personal assistant…"
+// boilerplate that filled the first ~500 chars, so their personality
+// markers (loaded from per-agent SOUL.md / IDENTITY.md / etc.) didn't
+// reach the hash and they collided onto the same session key.
+//
+// Bumping to 20kB covers realistic agent system prompts including
+// rich workspace bootstrap (Lux: ~42kB, Mercury: ~80kB total — but
+// the first 20kB has more than enough divergence to fingerprint each).
+// SHA-256 cost on 20kB is ~10-20µs, irrelevant per request.
+const SYSTEM_TRIM = 20000;
 const USER_TRIM = 500;
 
 /**

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mobygate",
-  "version": "0.8.1",
+  "version": "0.8.3",
   "description": "OpenAI-compatible local proxy for Claude Max. The Möbius-strip gateway: OpenAI shape in, Claude Max out.",
   "type": "module",
   "main": "server.js",

package/server.js

@@ -168,6 +168,17 @@ for (const sig of ['SIGTERM', 'SIGINT', 'SIGHUP']) {
 // Opus 4.7 ships a native 1M-context variant addressed as `claude-opus-4-7[1m]`.
 // Default opus aliases route to the 1M form to match the advertised context window.
 // Pass `claude-opus-4-7-200k` for the standard (cheaper) 200k variant.
+//
+// History: the sonnet-4-6 entry previously mapped to the dated
+// `claude-sonnet-4-5-20250929` because at the time, the SDK didn't
+// recognize `claude-sonnet-4-6` natively. The SDK has since added
+// native support for the un-dated 4-6 alias, so sonnet-4-6 requests
+// were silently being downgraded to retired 4-5-20250929. This caused
+// the "Sonnet only" Anthropic quota to show 0% usage even when Lux
+// and Mercury (configured for sonnet-4-6) were chatting actively —
+// the SDK was accepting the retired model id but Claude was likely
+// falling back to opus or returning a zero-billed response. Fixed in
+// v0.8.2 by routing 4-6 through directly.
 const MODEL_MAP = {
   'claude-opus-4': 'claude-opus-4-7[1m]',
   'claude-opus-4-6': 'claude-opus-4-6',
@@ -175,13 +186,14 @@ const MODEL_MAP = {
   'claude-opus-4-7[1m]': 'claude-opus-4-7[1m]',
   'claude-opus-4-7-1m': 'claude-opus-4-7[1m]',
   'claude-opus-4-7-200k': 'claude-opus-4-7',
-  'claude-sonnet-4': 'claude-sonnet-4-5-20250929',
-  'claude-sonnet-4-5': 'claude-sonnet-4-5-20250929',
-  'claude-sonnet-4-6': 'claude-sonnet-4-5-20250929', // SDK resolves 4-6 to non-existent dated version
+  'claude-sonnet-4': 'claude-sonnet-4-6[1m]',         // current latest sonnet, 1M context
+  'claude-sonnet-4-5': 'claude-sonnet-4-5-20250929',  // explicit request for older 4-5
+  'claude-sonnet-4-6': 'claude-sonnet-4-6[1m]',       // SDK supports natively; [1m] unlocks 1M context (same pattern as opus 4.7)
+  'claude-sonnet-4-6-200k': 'claude-sonnet-4-6',      // explicit 200k variant
   'claude-haiku-4': 'claude-haiku-4-5-20251001',
   'claude-haiku-4-5': 'claude-haiku-4-5-20251001',
   'opus': 'claude-opus-4-7[1m]',
-  'sonnet': 'claude-sonnet-4-5-20250929',
+  'sonnet': 'claude-sonnet-4-6[1m]',                  // current latest sonnet, 1M context
   'haiku': 'claude-haiku-4-5-20251001',
 };
 
@@ -579,6 +591,7 @@ async function handleStreaming(req, res, body, requestId, sessionKey) {
         outputTokens = usage.output_tokens;
         cacheReadTokens = usage.cache_read_input_tokens;
         cacheCreateTokens = usage.cache_creation_input_tokens;
+        console.log(`  [model-billed] requested=${resolvedModel} modelUsage=${JSON.stringify(usage.modelUsage || '(none)')}`);
         break;
       }
     }
@@ -784,6 +797,7 @@ async function handleNonStreaming(res, body, requestId, sessionKey) {
         outputTokens = usage.output_tokens;
         cacheReadTokens = usage.cache_read_input_tokens;
         cacheCreateTokens = usage.cache_creation_input_tokens;
+        console.log(`  [model-billed] requested=${resolvedModel} modelUsage=${JSON.stringify(usage.modelUsage || '(none)')}`);
         if (message.subtype) stopReason = message.subtype;
         break;
       }
@@ -998,6 +1012,7 @@ async function handleAnthropicNonStreaming(res, body, requestId, sessionKey) {
         outputTokens = usage.output_tokens;
         cacheReadTokens = usage.cache_read_input_tokens;
         cacheCreateTokens = usage.cache_creation_input_tokens;
+        console.log(`  [model-billed] requested=${resolvedModel} modelUsage=${JSON.stringify(usage.modelUsage || '(none)')}`);
         stopReason = mapStopReason(message);
         break;
       }
@@ -1243,6 +1258,7 @@ async function handleAnthropicStreaming(req, res, body, requestId, sessionKey) {
         outputTokens = usage.output_tokens;
         cacheReadTokens = usage.cache_read_input_tokens;
         cacheCreateTokens = usage.cache_creation_input_tokens;
+        console.log(`  [model-billed] requested=${resolvedModel} modelUsage=${JSON.stringify(usage.modelUsage || '(none)')}`);
         if (!toolUseEmitted) stopReason = mapStopReason(message);
         break;
       }