@1presence/bridge 0.44.0 → 0.46.0

package/dist/claude.js

@@ -161,6 +161,27 @@ function describeCliFailure(apiErrorText, authFailure) {
     }
     return 'Local Mode stopped unexpectedly. Please try again.';
 }
+/**
+ * Copy for an actionable rate-limit notice. The SDK emits `rate_limit_event`
+ * whenever rate-limit info CHANGES — including the routine `allowed` case on
+ * (nearly) every turn — so the caller must drop `allowed` and only invoke this
+ * for `allowed_warning` / `rejected`. `resetsAt` is a Unix timestamp; the SDK
+ * uses seconds, but we accept ms too in case that changes upstream.
+ */
+function formatRateLimitNotice(status, resetsAt) {
+    const when = typeof resetsAt === 'number' && resetsAt > 0
+        ? new Date(resetsAt < 1e12 ? resetsAt * 1000 : resetsAt)
+            .toLocaleTimeString([], { hour: 'numeric', minute: '2-digit' })
+        : null;
+    if (status === 'rejected') {
+        return when
+            ? `Pausing for an upstream rate limit — resumes around ${when}.`
+            : 'Pausing briefly for an upstream rate limit, then continuing…';
+    }
+    return when
+        ? `Approaching your usage limit — window resets around ${when}.`
+        : 'Approaching your usage limit for this period.';
+}
 // ─── Prompt construction ─────────────────────────────────────────────────────────
 //
 // The gateway pushes the FULL conversation (sanitised via @presence/shared
@@ -292,6 +313,27 @@ export function spawnClaude(params) {
     // credentials in the Keychain), not an API key that would bill a separate
     // account. Options.env REPLACES the subprocess env, so spread the rest through.
     const { ANTHROPIC_API_KEY: _stripped, ...safeEnv } = process.env;
+    // Claude Code truncates "large" MCP tool results (token count over
+    // MAX_MCP_OUTPUT_TOKENS, default 25,000) one of two ways, gated by
+    // ENABLE_MCP_LARGE_OUTPUT_FILES:
+    //   • default (feature ON): the full result is written to a tool-results/*.json
+    //     file and the model is handed a `<persisted-output>` stub telling it to Read
+    //     the file. Local Mode disables every built-in tool (`tools: []` below), so
+    //     the model has no Read — it reaches for vault_read, which resolves the local
+    //     path against the GCS prefix and fails ("No such object"). Unrecoverable.
+    //   • OFF: the result is inline-truncated to the cap — graceful degradation the
+    //     model can actually work with.
+    // Hosted mode pipes results straight to the Anthropic SDK with no truncation, so
+    // this breaks Local Mode only — a parity bug. Fix on two fronts: force the
+    // file-save path off so the unrecoverable failure mode is impossible, and raise
+    // the cap so legitimate large results (skill bodies, vault reads) pass inline
+    // instead of truncating. Operator overrides (env already set) win on both.
+    if (!safeEnv['ENABLE_MCP_LARGE_OUTPUT_FILES']) {
+        safeEnv['ENABLE_MCP_LARGE_OUTPUT_FILES'] = 'false';
+    }
+    if (!safeEnv['MAX_MCP_OUTPUT_TOKENS']) {
+        safeEnv['MAX_MCP_OUTPUT_TOKENS'] = '200000';
+    }
     const pinnedModel = getBridgeModel();
     // Process one translated raw stream-json event: bookkeeping + forward. Mirrors
     // the old CLI stdout parser so the gateway/accumulator see identical shapes.
@@ -557,9 +599,16 @@ export function spawnClaude(params) {
                         break;
                     }
                     case 'rate_limit_event': {
-                        // SDK surfaces upstream rate-limit pauses; it retries internally.
-                        // Admin-only ephemeral notice — jargon is fine in Local Mode.
-                        onNotice?.('Claude Code is pausing briefly for an upstream rate limit, then continuing…');
+                        // The SDK emits this whenever rate-limit info CHANGES — including the
+                        // routine `allowed` case on (nearly) every turn, which previously
+                        // spammed a phantom "pausing" notice (see vault/Bugs.md). Only surface
+                        // a notice when the user is actually warned or throttled.
+                        const info = m.rate_limit_info;
+                        const status = info?.status;
+                        if (status === 'allowed_warning' || status === 'rejected') {
+                            // Admin-only ephemeral notice — jargon is fine in Local Mode.
+                            onNotice?.(formatRateLimitNotice(status, info?.resetsAt));
+                        }
                         break;
                     }
                     // Everything else (partial messages, status, hooks, notifications,

package/dist/index.js

@@ -327,6 +327,8 @@ async function handleMessage(conversationId, text, sessionId, history, auth, vau
             // Ephemeral, non-persisted thread notice (admin-only Local Mode). Relayed
             // by the gateway to the PWA SSE stream as a `notice` AgentEvent; it does
             // NOT go through the turn accumulator, so it never lands in history.
+            // Log it too: anything the user sees in chat must have a bridge-log trail.
+            console.log(`[${new Date().toLocaleTimeString()}] ⚠ notice: ${message}`);
             if (currentWs?.readyState === WebSocket.OPEN) {
                 currentWs.send(JSON.stringify({ type: 'notice', conversationId, message }));
             }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@1presence/bridge",
-  "version": "0.44.0",
+  "version": "0.46.0",
   "description": "Run 1Presence on your Mac and use your Claude.ai Pro subscription from any device",
   "type": "module",
   "bin": {