fix: truncate large tool outputs to prevent Crof incomplete responses

Crof models (mimo, deepseek-v4-pro) return status=incomplete when
tool results contain too much text (e.g. full HTML pages at 8500+ tokens).
Auto-truncate tool outputs exceeding 8000 chars with truncation notice.
Combined with the 30-item conversation trim from previous commit.

CHANGELOG.md

@@ -9,6 +9,9 @@
   - Keeps system/developer messages, original user query, and most recent items
   - Drops oldest tool call/outputs from the middle when conversation grows too long
   - Prevents `status=incomplete` errors on providers with smaller context windows
+- **Truncates large tool outputs (>8000 chars)** to prevent model output token exhaustion
+  - Crof's models return `incomplete` when tool results contain too much text (e.g., full HTML pages)
+  - Truncated outputs include `[truncated N chars]` suffix so the model knows data was cut
 - Added request/response logging to `~/.cache/codex-proxy/requests.log` for debugging
 - Proxy stderr no longer discarded by launcher (visible in terminal for debugging)
 

src/translate-proxy.py

@@ -166,12 +166,24 @@ def forwarded_headers(request_headers, extra=None, browser_ua=False):
     return headers
 
 _MAX_INPUT_ITEMS = 30
+_MAX_TOOL_OUTPUT_CHARS = 8000
 
 def _trim_input(input_data):
-    if not isinstance(input_data, list) or len(input_data) <= _MAX_INPUT_ITEMS:
+    if not isinstance(input_data, list):
         return input_data
+    out = []
+    for item in input_data:
+        if item.get("type") == "function_call_output":
+            o = item.get("output", "")
+            if len(o) > _MAX_TOOL_OUTPUT_CHARS:
+                item = dict(item)
+                item["output"] = o[:_MAX_TOOL_OUTPUT_CHARS] + f"\n... [truncated {len(o) - _MAX_TOOL_OUTPUT_CHARS} chars]"
+                print(f"[trim] tool output truncated {len(o)} -> {_MAX_TOOL_OUTPUT_CHARS}", file=sys.stderr)
+        out.append(item)
+    if len(out) <= _MAX_INPUT_ITEMS:
+        return out
     head_end = 0
-    for i, item in enumerate(input_data):
+    for i, item in enumerate(out):
         t = item.get("type")
         if t == "message" and item.get("role") in ("developer", "system"):
             head_end = i + 1
@@ -179,12 +191,12 @@ def _trim_input(input_data):
             head_end = i + 1
         else:
             break
-    head = input_data[:head_end]
+    head = out[:head_end]
     tail_keep = _MAX_INPUT_ITEMS - len(head)
-    tail = input_data[-tail_keep:]
-    trimmed = len(input_data) - len(head) - len(tail)
+    tail = out[-tail_keep:]
+    trimmed = len(out) - len(head) - len(tail)
     if trimmed > 0:
-        print(f"[trim] {len(input_data)} items -> {len(head) + len(tail)} (dropped {trimmed} old items)", file=sys.stderr)
+        print(f"[trim] {len(out)} items -> {len(head) + len(tail)} (dropped {trimmed} old items)", file=sys.stderr)
     return head + tail
 
 # ═══════════════════════════════════════════════════════════════════