@elvatis_com/openclaw-cli-bridge-elvatis 2.10.0 → 2.10.1

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:** `2.10.0`
+**Current version:** `2.10.1`
 
 ---
 
@@ -406,6 +406,11 @@ npm run ci          # lint + typecheck + test
 
 ## Changelog
 
+### v2.10.1
+- **feat:** smart tool-routing — tool-heavy requests (>8 tools) auto-route to Haiku instead of Sonnet. Haiku handles tool calls in ~11s vs Sonnet's 80-120s (with intermittent hangs). Sonnet is preserved for reasoning/text responses.
+- **fix:** reduce stale-output timeout 120s→60s — faster fallback when Sonnet goes silent
+- **feat:** per-model spawn logging with prompt size for debugging
+
 ### v2.10.0
 - **fix:** cap effective timeout at 580s (under gateway's 600s `idleTimeoutSeconds`) so bridge fallback fires BEFORE gateway kills the request — eliminates the race condition where both compete to handle the timeout
 - **fix:** reduce Sonnet base timeout 420s→300s, Opus 420s→360s — ensures fallback triggers faster for stuck CLI sessions

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:** 2.10.0
+**Version:** 2.10.1

package/openclaw.plugin.json

@@ -2,7 +2,7 @@
   "id": "openclaw-cli-bridge-elvatis",
   "slug": "openclaw-cli-bridge-elvatis",
   "name": "OpenClaw CLI Bridge",
-  "version": "2.10.0",
+  "version": "2.10.1",
   "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": "2.10.0",
+  "version": "2.10.1",
   "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

@@ -566,6 +566,7 @@ export async function runClaude(
     : prompt;
 
   const cwd = workdir ?? homedir();
+  debugLog("CLAUDE", `spawn ${model}`, { promptLen: effectivePrompt.length, promptKB: Math.round(effectivePrompt.length / 1024), cwd, timeoutMs: Math.round(timeoutMs / 1000) });
   const result = await runCli("claude", args, effectivePrompt, timeoutMs, { cwd, log: opts?.log });
 
   // On 401: attempt one token refresh + retry before giving up.

package/src/config.ts

@@ -57,7 +57,7 @@ export const TIMEOUT_GRACE_MS = 5_000;
  * assume it's stuck and SIGTERM early. 0 = disabled.
  * Prevents waiting the full timeout when Claude CLI hangs silently.
  */
-export const STALE_OUTPUT_TIMEOUT_MS = 120_000; // 2 min of silence → kill
+export const STALE_OUTPUT_TIMEOUT_MS = 60_000; // 1 min of silence → kill (Sonnet goes silent when rate-limited)
 
 /** Max messages to include in the prompt sent to CLI subprocesses. */
 export const MAX_MESSAGES = 20;
@@ -71,6 +71,13 @@ export const MAX_MESSAGES_HEAVY_TOOLS = 12;
 /** Tool count threshold that triggers reduced message limit. */
 export const TOOL_HEAVY_THRESHOLD = 10;
 
+/**
+ * Tool count threshold that triggers smart routing to a faster model.
+ * When Sonnet receives a request with this many tools, route to Haiku instead.
+ * Haiku handles tool calls in ~11s vs Sonnet's 80-120s (and Sonnet hangs intermittently).
+ */
+export const TOOL_ROUTING_THRESHOLD = 8;
+
 /** Max characters per message content before truncation. */
 export const MAX_MSG_CHARS = 4_000;
 

package/src/proxy-server.ts

@@ -32,6 +32,7 @@ import {
   BITNET_MAX_MESSAGES,
   BITNET_SYSTEM_PROMPT,
   DEFAULT_MODEL_TIMEOUTS,
+  TOOL_ROUTING_THRESHOLD,
 } from "./config.js";
 import { debugLog, DEBUG_LOG_PATH } from "./debug-log.js";
 
@@ -790,6 +791,18 @@ async function handleRequest(
     // ── CLI runner routing (Gemini / Claude Code / Codex) ──────────────────────
     let result: CliToolResult;
     let usedModel = model;
+
+    // ── Smart tool routing: heavy tool requests → Haiku for speed ──────────
+    // Sonnet hangs intermittently on large tool prompts (20KB+, 21 tools).
+    // Haiku handles tool calls in ~11s vs Sonnet's 80-120s (when it works).
+    // Route tool-heavy requests directly to Haiku, keep Sonnet for reasoning.
+    if (hasTools && tools!.length > TOOL_ROUTING_THRESHOLD && model === "cli-claude/claude-sonnet-4-6") {
+      const toolModel = "cli-claude/claude-haiku-4-5";
+      opts.log(`[cli-bridge] tool-routing: ${model} → ${toolModel} (${tools!.length} tools)`);
+      debugLog("TOOL-ROUTE", `${model} → ${toolModel}`, { tools: tools!.length, threshold: TOOL_ROUTING_THRESHOLD });
+      usedModel = toolModel;
+    }
+
     const routeOpts = { workdir, tools: hasTools ? tools : undefined, mediaFiles: mediaFiles.length ? mediaFiles : undefined, log: opts.log };
 
     // ── Provider session: ensure a persistent session for this model ────────
@@ -843,12 +856,12 @@ async function handleRequest(
 
     const cliStart = Date.now();
     try {
-      result = await routeToCliRunner(model, cleanMessages, effectiveTimeout, routeOpts);
+      result = await routeToCliRunner(usedModel, cleanMessages, effectiveTimeout, routeOpts);
       const latencyMs = Date.now() - cliStart;
       const estCompletionTokens = estimateTokens(result.content ?? "");
-      metrics.recordRequest(model, latencyMs, true, estPromptTokens, estCompletionTokens, promptPreview);
+      metrics.recordRequest(usedModel, latencyMs, true, estPromptTokens, estCompletionTokens, promptPreview);
       providerSessions.recordRun(session.id, false);
-      debugLog("OK", `${model} completed in ${(latencyMs / 1000).toFixed(1)}s`, { toolCalls: result.tool_calls?.length ?? 0, contentLen: result.content?.length ?? 0 });
+      debugLog("OK", `${usedModel} completed in ${(latencyMs / 1000).toFixed(1)}s`, { toolCalls: result.tool_calls?.length ?? 0, contentLen: result.content?.length ?? 0 });
     } catch (err) {
       const primaryDuration = Date.now() - cliStart;
       const msg = (err as Error).message;