@elvatis_com/openclaw-cli-bridge-elvatis 3.4.1 → 3.5.1

package/README.md

@@ -2,7 +2,7 @@
 
 > OpenClaw plugin that bridges locally installed AI CLIs (Codex, Gemini, Claude Code, OpenCode, Pi) as model providers — with slash commands for instant model switching, restore, health testing, and model listing.
 
-**Current version:** `3.4.1`
+**Current version:** `3.5.1`
 
 ---
 

package/SKILL.md

@@ -68,4 +68,4 @@ On gateway restart, if any session has expired, a **WhatsApp alert** is sent aut
 
 See `README.md` for full configuration reference and architecture diagram.
 
-**Version:** 3.4.1
+**Version:** 3.5.1

package/openclaw.plugin.json

@@ -2,7 +2,7 @@
   "id": "openclaw-cli-bridge-elvatis",
   "slug": "openclaw-cli-bridge-elvatis",
   "name": "OpenClaw CLI Bridge",
-  "version": "3.4.1",
+  "version": "3.5.1",
   "license": "MIT",
   "description": "Phase 1: openai-codex auth bridge. Phase 2: local HTTP proxy routing model calls through gemini/claude CLIs (vllm provider).",
   "providers": [

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@elvatis_com/openclaw-cli-bridge-elvatis",
-  "version": "3.4.1",
+  "version": "3.5.1",
   "description": "Bridges gemini, claude, and codex CLI tools as OpenClaw model providers. Reads existing CLI auth without re-login.",
   "type": "module",
   "openclaw": {

package/src/cli-runner.ts

@@ -80,11 +80,16 @@ export function formatPrompt(messages: ChatMessage[], toolCount = 0): string {
   // Reduce history when tool schemas dominate the prompt
   const maxMsgs = toolCount > TOOL_HEAVY_THRESHOLD ? MAX_MESSAGES_HEAVY_TOOLS : MAX_MESSAGES;
 
-  // Keep system message (if any) + last N non-system messages
+  // Keep system message (if any) + first user message (original request) + last N non-system messages
   const system = messages.find((m) => m.role === "system");
   const nonSystem = messages.filter((m) => m.role !== "system");
+  const firstUser = nonSystem.find((m) => m.role === "user");
   const recent = nonSystem.slice(-maxMsgs);
-  const truncated = system ? [system, ...recent] : recent;
+  // Pin the first user message so the model never loses the original request
+  const pinned = firstUser && !recent.includes(firstUser)
+    ? [firstUser, ...recent]
+    : recent;
+  const truncated = system ? [system, ...pinned] : pinned;
 
   // Single short user message — send bare (no wrapping needed)
   if (truncated.length === 1 && truncated[0].role === "user") {
@@ -1022,9 +1027,13 @@ export async function routeToCliRunner(
   let prompt = formatPrompt(messages, toolCount);
   const hasTools = toolCount > 0;
 
-  // Auto-detect project from prompt and set workdir + inject context
+  // Auto-detect project from user messages only (not tool results which mention other projects)
   if (!opts.workdir) {
-    const detected = detectProjectFromPrompt(prompt);
+    const userText = messages
+      .filter((m) => m.role === "user")
+      .map((m) => typeof m.content === "string" ? m.content : "")
+      .join(" ");
+    const detected = detectProjectFromPrompt(userText);
     if (detected) {
       opts = { ...opts, workdir: detected.path };
       prompt = `[Context: Working directory is ${detected.path}]\n\n${prompt}`;

package/src/proxy-server.ts

@@ -448,6 +448,9 @@ async function handleRequest(
     const promptPreview = typeof lastUserMsg?.content === "string" ? lastUserMsg.content.slice(0, 80) : "";
 
     debugLog("REQ", `${model} start`, { msgs: cleanMessages.length, tools: tools?.length ?? 0, stream, media: mediaFiles.length, promptPreview: promptPreview.slice(0, 60) });
+    if (hasTools && tools!.length > 0) {
+      debugLog("TOOLS", `${tools!.length} tools available`, { names: tools!.map(t => t.function?.name ?? t.name ?? "?").join(", ") });
+    }
 
     // Track active request for dashboard
     activeRequests.set(id, { id, model, startedAt: Date.now(), messageCount: cleanMessages.length, toolCount: tools?.length ?? 0, promptPreview });
@@ -939,7 +942,14 @@ async function handleRequest(
         opts.warn(`[cli-bridge] ${model} failed (${reason}), trying fallback chain: ${fallbackChain.join(" → ")}`);
 
         let chainSuccess = false;
+        const lastMsg = cleanMessages[cleanMessages.length - 1];
+        const inToolLoop = hasTools && (lastMsg?.role === "tool" || lastMsg?.role === "function");
         for (const fallbackModel of fallbackChain) {
+          // Skip Haiku in tool loops — it consistently returns text instead of tool_calls, wasting ~8-12s
+          if (inToolLoop && fallbackModel.includes("haiku")) {
+            debugLog("FALLBACK-SKIP", `skipping ${fallbackModel} in tool loop (unreliable for tool_calls)`, {});
+            continue;
+          }
           debugLog("FALLBACK", `${model} → ${fallbackModel}`, { reason: isTimeout ? "timeout" : "error", primaryDuration: Math.round(primaryDuration / 1000), chain: fallbackChain });
           if (sseHeadersSent) {
             res.write(`: fallback — trying ${fallbackModel}\n\n`);
@@ -952,12 +962,9 @@ async function handleRequest(
               debugLog("FALLBACK-EMPTY", `${fallbackModel} returned empty`, {});
               throw new Error(`empty response from ${fallbackModel}`);
             }
-            // If tools were requested and the last message was a tool result (gateway expects
-            // tool continuation), but the fallback model returned text instead of tool_calls —
+            // If we're in a tool loop but the fallback returned text instead of tool_calls —
             // it ignored the JSON format. Try next model in chain.
-            const lastMsg = cleanMessages[cleanMessages.length - 1];
-            const inToolLoop = lastMsg?.role === "tool" || lastMsg?.role === "function";
-            if (hasTools && inToolLoop && !result.tool_calls?.length && result.content) {
+            if (inToolLoop && !result.tool_calls?.length && result.content) {
               debugLog("FALLBACK-NO-TOOLS", `${fallbackModel} returned text instead of tool_calls in tool loop`, { contentLen: result.content.length, preview: result.content.slice(0, 80) });
               throw new Error(`${fallbackModel} returned text instead of tool_calls`);
             }

package/src/tool-protocol.ts

@@ -188,6 +188,16 @@ export function parseToolCallResponse(text: string): CliToolResult {
     }
   }
 
+  // Last resort: try to rescue tool_calls from anywhere in the text
+  // Models sometimes output tool_calls JSON with surrounding text that breaks other strategies
+  if (trimmed.includes("tool_calls")) {
+    const rescued = tryRescueToolCallsFromContent(trimmed);
+    if (rescued) {
+      debugLog("PARSE", `rescue-from-raw → tool_calls`, { toolCalls: rescued.tool_calls?.length ?? 0 });
+      return rescued;
+    }
+  }
+
   // Fallback: treat entire text as content
   debugLog("PARSE", "no JSON found → raw content", { len: trimmed.length, preview });
   return { content: trimmed || null };
@@ -307,37 +317,49 @@ function tryExtractCodeBlock(text: string): string | null {
   return match?.[1]?.trim() ?? null;
 }
 
-/** Find the first { ... } JSON object in text (greedy, balanced braces). */
+/** Find a balanced { ... } JSON object in text. Tries multiple start positions if the first fails to parse. */
 function tryExtractEmbeddedJson(text: string): string | null {
-  const start = text.indexOf("{");
-  if (start === -1) return null;
-
-  let depth = 0;
-  let inString = false;
-  let escaped = false;
-
-  for (let i = start; i < text.length; i++) {
-    const ch = text[i];
-    if (escaped) {
-      escaped = false;
-      continue;
-    }
-    if (ch === "\\") {
-      escaped = true;
-      continue;
-    }
-    if (ch === '"') {
-      inString = !inString;
-      continue;
-    }
-    if (inString) continue;
-    if (ch === "{") depth++;
-    if (ch === "}") {
-      depth--;
-      if (depth === 0) {
-        return text.slice(start, i + 1);
+  let searchFrom = 0;
+  while (searchFrom < text.length) {
+    const start = text.indexOf("{", searchFrom);
+    if (start === -1) return null;
+
+    let depth = 0;
+    let inString = false;
+    let escaped = false;
+
+    for (let i = start; i < text.length; i++) {
+      const ch = text[i];
+      if (escaped) {
+        escaped = false;
+        continue;
+      }
+      if (ch === "\\") {
+        escaped = true;
+        continue;
+      }
+      if (ch === '"') {
+        inString = !inString;
+        continue;
+      }
+      if (inString) continue;
+      if (ch === "{") depth++;
+      if (ch === "}") {
+        depth--;
+        if (depth === 0) {
+          const candidate = text.slice(start, i + 1);
+          // Verify it actually parses as JSON before returning
+          try {
+            JSON.parse(candidate);
+            return candidate;
+          } catch {
+            // This balanced-brace block isn't valid JSON — try next { in text
+            break;
+          }
+        }
       }
     }
+    searchFrom = start + 1;
   }
   return null;
 }

package/test/cli-runner.test.ts

@@ -19,18 +19,19 @@ describe("formatPrompt", () => {
     expect(result).toBe("hello");
   });
 
-  it("truncates to MAX_MESSAGES (20) non-system messages", () => {
+  it("truncates to MAX_MESSAGES (20) non-system messages but pins first user message", () => {
     const messages = Array.from({ length: 30 }, (_, i) => ({
       role: "user" as const,
       content: `msg ${i}`,
     }));
     const result = formatPrompt(messages);
     expect(result).toContain("msg 29");
-    expect(result).not.toContain("msg 0\n");
+    expect(result).toContain("msg 0"); // first user message is always pinned
+    expect(result).not.toContain("msg 1\n"); // but intermediate messages are truncated
     expect(result).toContain("[User]");
   });
 
-  it("keeps system message + last 20 non-system messages", () => {
+  it("keeps system message + first user message + last 20 non-system messages", () => {
     const sys = { role: "system" as const, content: "You are helpful" };
     const msgs = Array.from({ length: 25 }, (_, i) => ({
       role: "user" as const,
@@ -40,7 +41,8 @@ describe("formatPrompt", () => {
     expect(result).toContain("[System]");
     expect(result).toContain("You are helpful");
     expect(result).toContain("msg 24");
-    expect(result).not.toContain("msg 0\n");
+    expect(result).toContain("msg 0"); // first user message is always pinned
+    expect(result).not.toContain("msg 1\n"); // but intermediate messages are truncated
   });
 
   it("truncates individual message content at MAX_MSG_CHARS (4000)", () => {