Compare commits
5 Commits
26
CHANGELOG.md
26
CHANGELOG.md
@@ -1,5 +1,31 @@
|
||||
# Changelog
|
||||
|
||||
## v3.11.12 (2026-05-26)
|
||||
|
||||
**New Antigravity v2 Handler (Mimicking anti-api)**
|
||||
|
||||
### New Features
|
||||
- **Complete rewrite of Antigravity handler** based on https://github.com/ink1ing/anti-api approach
|
||||
- Safety settings (all OFF), stopSequences, sessionId, requestType: agent
|
||||
- functionResponse uses `response: { result: string }` format matching anti-api
|
||||
- Endpoint priority: `daily-cloudcode-pa.googleapis.com` first
|
||||
- Simplified sanitizer: only deduplicates consecutive user text, never touches tool messages
|
||||
|
||||
## v3.11.11 (2026-05-26)
|
||||
|
||||
## v3.11.11 (2026-05-26)
|
||||
|
||||
**Antigravity Fix: Stricter function_call/output Pairing + Gemini Sanitizer Rewrite (PR #12)**
|
||||
|
||||
### Bug Fixes
|
||||
- **Stricter function_call/output pairing**: Only includes pairs where BOTH call and output exist — no orphan calls sent to Gemini
|
||||
- **Gemini sanitizer rewritten**: Tool messages (`functionCall`/`functionResponse`) are always preserved as-is, never merged or skipped
|
||||
- **Text merging more conservative**: Checks last message for tool content before merging consecutive text messages
|
||||
- **Final trimming safe**: Only removes plain `message` items, never `function_call_output` (which would break tool pairs)
|
||||
- **Merge PR #12**: Fix by qwen-chat coder
|
||||
|
||||
## v3.11.10 (2026-05-26)
|
||||
|
||||
## v3.11.10 (2026-05-26)
|
||||
|
||||
**Antigravity Fix: Interleave function_call/output Pairs, Gemini Turn Trimming (PR #11)**
|
||||
|
||||
Binary file not shown.
BIN
codex-launcher_3.11.12_all.deb
Normal file
BIN
codex-launcher_3.11.12_all.deb
Normal file
Binary file not shown.
@@ -27,6 +27,10 @@ model_catalog_json = ""
|
||||
"""
|
||||
|
||||
CHANGELOG = [
|
||||
("3.11.12", "2026-05-26", [
|
||||
"New Antigravity v2 handler mimicking anti-api",
|
||||
"Safety settings, stopSequences, simplified sanitizer",
|
||||
]),
|
||||
("3.11.10", "2026-05-26", [
|
||||
"Fix Antigravity: interleave function_call/output pairs (PR #11)",
|
||||
"Gemini sanitizer: trim non-user turns for Google API compliance",
|
||||
|
||||
@@ -83,6 +83,16 @@ model_catalog_json = ""
|
||||
"""
|
||||
|
||||
CHANGELOG = [
|
||||
("3.11.12", "2026-05-26", [
|
||||
"New Antigravity v2 handler mimicking anti-api approach",
|
||||
"Safety settings, stopSequences, sessionId, requestType: agent",
|
||||
"Simplified sanitizer preserving functionCall/functionResponse",
|
||||
"Endpoint priority: daily-cloudcode-pa first",
|
||||
"functionResponse uses response.result (string) format",
|
||||
]),
|
||||
("3.11.11", "2026-05-26", [
|
||||
"Final trimming only removes plain messages, never function_call_output",
|
||||
]),
|
||||
("3.11.10", "2026-05-26", [
|
||||
"Fix Antigravity: interleave function_call/output pairs in correct sequence (PR #11)",
|
||||
"Fix Gemini sanitizer: trim leading/trailing non-user turns for Google API compliance",
|
||||
|
||||
@@ -320,9 +320,10 @@ _active_requests = {}
|
||||
_active_requests_lock = threading.Lock()
|
||||
|
||||
_pool = uuid.uuid4().hex[:8]
|
||||
_antigravity_version = "1.18.3"
|
||||
_antigravity_version = "2.0.1"
|
||||
_antigravity_version_checked = 0
|
||||
_antigravity_version_lock = threading.Lock()
|
||||
_antigravity_version_validated = False
|
||||
_last_user_urls = collections.deque(maxlen=20)
|
||||
|
||||
_conn_pool_lock = threading.Lock()
|
||||
@@ -798,49 +799,137 @@ _ANTIGRAVITY_LOOP_TRACKER_LOCK = threading.Lock()
|
||||
def _antigravity_loop_key(session_id):
|
||||
return f"ag:{session_id}"
|
||||
|
||||
def _validate_antigravity_version(version, access_token=None, project_id=None):
|
||||
if not version or not re.match(r"^\d+\.\d+\.\d+$", version):
|
||||
return False
|
||||
try:
|
||||
if not access_token:
|
||||
access_token = _refresh_oauth_token()
|
||||
if not project_id:
|
||||
token_path = os.path.join(os.path.expanduser("~"), ".cache", "codex-proxy", "google-antigravity-oauth-token.json")
|
||||
try:
|
||||
with open(token_path) as f:
|
||||
project_id = json.load(f).get("project_id", "")
|
||||
except Exception:
|
||||
pass
|
||||
if not access_token or not project_id:
|
||||
return True
|
||||
import platform as _plat
|
||||
_os_name = _plat.system().lower()
|
||||
_os_arch = _plat.machine().lower().replace("x86_64", "x64").replace("aarch64", "arm64")
|
||||
ua = f"antigravity/{version} {_os_name}/{_os_arch}"
|
||||
body = {
|
||||
"project": project_id,
|
||||
"model": "gemini-3-flash",
|
||||
"requestType": "agent",
|
||||
"userAgent": ua,
|
||||
"requestId": f"probe-{uuid.uuid4().hex[:8]}",
|
||||
"request": {
|
||||
"contents": [{"role": "user", "parts": [{"text": "hi"}]}],
|
||||
"sessionId": f"probe{int(time.time()*1000)}",
|
||||
"safetySettings": [
|
||||
{"category": "HARM_CATEGORY_HARASSMENT", "threshold": "OFF"},
|
||||
{"category": "HARM_CATEGORY_HATE_SPEECH", "threshold": "OFF"},
|
||||
{"category": "HARM_CATEGORY_SEXUALLY_EXPLICIT", "threshold": "OFF"},
|
||||
{"category": "HARM_CATEGORY_DANGEROUS_CONTENT", "threshold": "OFF"},
|
||||
{"category": "HARM_CATEGORY_CIVIC_INTEGRITY", "threshold": "OFF"},
|
||||
],
|
||||
"generationConfig": {"maxOutputTokens": 32, "stopSequences": ["\n\nHuman:", "[DONE]"]},
|
||||
}
|
||||
}
|
||||
url = "https://daily-cloudcode-pa.googleapis.com/v1internal:streamGenerateContent?alt=sse"
|
||||
headers = {
|
||||
"Content-Type": "application/json",
|
||||
"Authorization": f"Bearer {access_token}",
|
||||
"User-Agent": ua,
|
||||
}
|
||||
req = urllib.request.Request(url, data=json.dumps(body).encode(), headers=headers)
|
||||
resp = urllib.request.urlopen(req, timeout=15)
|
||||
data = resp.read().decode()
|
||||
if "no longer supported" in data.lower():
|
||||
print(f"[antigravity-version] version {version} rejected (deprecated)", file=sys.stderr)
|
||||
return False
|
||||
return True
|
||||
except urllib.error.HTTPError as e:
|
||||
if e.code == 404:
|
||||
print(f"[antigravity-version] version {version} rejected (404)", file=sys.stderr)
|
||||
return False
|
||||
return True
|
||||
except Exception as e:
|
||||
print(f"[antigravity-version] probe error for {version}: {e}", file=sys.stderr)
|
||||
return True
|
||||
|
||||
def _fetch_antigravity_version():
|
||||
cache_path = os.path.join(os.path.expanduser("~"), ".cache", "codex-proxy", "antigravity-version.json")
|
||||
try:
|
||||
with open(cache_path) as f:
|
||||
cached = json.load(f)
|
||||
if cached.get("version") and cached.get("checked_at", 0) > time.time() - 6 * 3600:
|
||||
if cached.get("version") and cached.get("validated") and cached.get("checked_at", 0) > time.time() - 6 * 3600:
|
||||
return cached["version"]
|
||||
except Exception:
|
||||
pass
|
||||
urls = [
|
||||
|
||||
access_token = None
|
||||
project_id = None
|
||||
try:
|
||||
access_token = _refresh_oauth_token()
|
||||
token_path = os.path.join(os.path.expanduser("~"), ".cache", "codex-proxy", "google-antigravity-oauth-token.json")
|
||||
with open(token_path) as f:
|
||||
project_id = json.load(f).get("project_id", "")
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
sources = [
|
||||
("https://antigravity-auto-updater-974169037036.us-central1.run.app", None),
|
||||
("https://antigravity.google/changelog", 5000),
|
||||
]
|
||||
for url, limit in urls:
|
||||
|
||||
candidates = []
|
||||
for url, limit in sources:
|
||||
try:
|
||||
req = urllib.request.Request(url, headers={"User-Agent": "Mozilla/5.0"})
|
||||
resp = urllib.request.urlopen(req, timeout=5)
|
||||
text = resp.read().decode(errors="replace")
|
||||
if limit:
|
||||
text = text[:limit]
|
||||
m = re.search(r"\d+\.\d+\.\d+", text)
|
||||
if m:
|
||||
version = m.group(0)
|
||||
for m in re.finditer(r"\d+\.\d+\.\d+", text):
|
||||
ver = m.group(0)
|
||||
if ver not in candidates:
|
||||
candidates.append(ver)
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
for ver in candidates:
|
||||
if _validate_antigravity_version(ver, access_token, project_id):
|
||||
print(f"[antigravity-version] fetched version {ver} validated", file=sys.stderr)
|
||||
try:
|
||||
os.makedirs(os.path.dirname(cache_path), exist_ok=True)
|
||||
with open(cache_path, "w", encoding="utf-8") as f:
|
||||
json.dump({"version": version, "checked_at": time.time()}, f)
|
||||
json.dump({"version": ver, "validated": True, "checked_at": time.time()}, f)
|
||||
except Exception:
|
||||
pass
|
||||
return version
|
||||
return ver
|
||||
|
||||
fallback = "2.0.1"
|
||||
print(f"[antigravity-version] all candidates failed, using fallback {fallback}", file=sys.stderr)
|
||||
try:
|
||||
os.makedirs(os.path.dirname(cache_path), exist_ok=True)
|
||||
with open(cache_path, "w", encoding="utf-8") as f:
|
||||
json.dump({"version": fallback, "validated": False, "checked_at": time.time()}, f)
|
||||
except Exception:
|
||||
pass
|
||||
return _antigravity_version
|
||||
return fallback
|
||||
|
||||
def _ensure_antigravity_version():
|
||||
global _antigravity_version, _antigravity_version_checked
|
||||
if time.time() - _antigravity_version_checked < 6 * 3600:
|
||||
global _antigravity_version, _antigravity_version_checked, _antigravity_version_validated
|
||||
if _antigravity_version_validated and time.time() - _antigravity_version_checked < 6 * 3600:
|
||||
return _antigravity_version
|
||||
with _antigravity_version_lock:
|
||||
if time.time() - _antigravity_version_checked < 6 * 3600:
|
||||
if _antigravity_version_validated and time.time() - _antigravity_version_checked < 6 * 3600:
|
||||
return _antigravity_version
|
||||
_antigravity_version = _fetch_antigravity_version()
|
||||
_antigravity_version_checked = time.time()
|
||||
_antigravity_version_validated = True
|
||||
return _antigravity_version
|
||||
|
||||
_antigravity_client_version = "1.110.0"
|
||||
@@ -4841,6 +4930,14 @@ def _antigravity_is_simple_user(text):
|
||||
return False
|
||||
|
||||
def _antigravity_normalize_context(input_data, model=""):
|
||||
"""
|
||||
Normalize context for Antigravity while PRESERVING function_call -> function_call_output pairs.
|
||||
|
||||
Google's Gemini API requires STRICT alternation:
|
||||
- functionCall (role=model) MUST be immediately followed by functionResponse (role=user)
|
||||
|
||||
This function compacts old history but NEVER breaks tool call/response pairs.
|
||||
"""
|
||||
if not isinstance(input_data, list) or len(input_data) < 2:
|
||||
return input_data
|
||||
is_claude_model = "claude" in model.lower()
|
||||
@@ -4889,7 +4986,7 @@ def _antigravity_normalize_context(input_data, model=""):
|
||||
dev_messages = []
|
||||
recent_items = []
|
||||
tool_outputs = []
|
||||
other_items = []
|
||||
tool_calls = []
|
||||
|
||||
for i, item in enumerate(input_data):
|
||||
if not isinstance(item, dict):
|
||||
@@ -4899,8 +4996,8 @@ def _antigravity_normalize_context(input_data, model=""):
|
||||
dev_messages.append(item)
|
||||
elif t == "function_call_output":
|
||||
tool_outputs.append((i, item))
|
||||
elif t in ("function_call",):
|
||||
other_items.append((i, item))
|
||||
elif t == "function_call":
|
||||
tool_calls.append((i, item))
|
||||
elif t == "message":
|
||||
recent_items.append((i, item))
|
||||
|
||||
@@ -4946,19 +5043,14 @@ def _antigravity_normalize_context(input_data, model=""):
|
||||
deduped_tail.append((idx, msg_item))
|
||||
recent_tail = deduped_tail if deduped_tail else recent_tail
|
||||
|
||||
tool_call_ids = set()
|
||||
for _, t_item in kept_tools:
|
||||
cid = t_item.get("call_id", t_item.get("id", ""))
|
||||
# Build call_id -> function_call mapping
|
||||
tool_call_map = {}
|
||||
for _, call_item in tool_calls:
|
||||
cid = call_item.get("call_id", call_item.get("id", ""))
|
||||
if cid:
|
||||
tool_call_ids.add(cid)
|
||||
tool_call_map[cid] = call_item
|
||||
|
||||
paired_calls = []
|
||||
for idx, item in other_items:
|
||||
cid = item.get("call_id", item.get("id", ""))
|
||||
if cid in tool_call_ids:
|
||||
paired_calls.append((idx, item))
|
||||
|
||||
# Build result maintaining proper function_call -> function_call_output pairing
|
||||
# Build result: maintain PAIRED sequence (function_call -> function_call_output)
|
||||
result = list(dev_messages)
|
||||
|
||||
compaction_summaries = []
|
||||
@@ -4974,29 +5066,21 @@ def _antigravity_normalize_context(input_data, model=""):
|
||||
summary_text = f"[Tool history summary: {n_summarized} older tool outputs omitted. {n_tool_calls} prior function calls were made for file inspection/editing.]"
|
||||
result.append({"type": "message", "role": "user", "content": [{"type": "input_text", "text": summary_text}]})
|
||||
|
||||
# CRITICAL FIX: Interleave function_calls with their corresponding function_call_outputs
|
||||
# to maintain the required sequence: function_call -> function_call_output -> function_call -> ...
|
||||
# Build a lookup map: call_id -> function_call_output item
|
||||
tool_output_map = {}
|
||||
# CRITICAL: Add tool CALLS and their corresponding OUTPUTS in PAIRED ORDER
|
||||
# Only include pairs where BOTH call and output are present
|
||||
added_pairs = set()
|
||||
for _, tool_item in kept_tools:
|
||||
cid = tool_item.get("call_id", tool_item.get("id", ""))
|
||||
if cid:
|
||||
tool_output_map[cid] = tool_item
|
||||
if cid and cid in tool_call_map and cid not in added_pairs:
|
||||
# Add function_call FIRST, then function_call_output IMMEDIATELY
|
||||
result.append(tool_call_map[cid])
|
||||
result.append(tool_item)
|
||||
added_pairs.add(cid)
|
||||
|
||||
# First, add all paired function_calls followed immediately by their responses
|
||||
added_call_ids = set()
|
||||
for _, call_item in paired_calls:
|
||||
cid = call_item.get("call_id", call_item.get("id", ""))
|
||||
result.append(call_item)
|
||||
added_call_ids.add(cid)
|
||||
# Immediately append the corresponding function_call_output if available
|
||||
if cid in tool_output_map:
|
||||
result.append(tool_output_map[cid])
|
||||
|
||||
# Add any remaining tool outputs that weren't paired (orphans)
|
||||
# Add any orphan tool outputs (no matching call found) - these go at the end before messages
|
||||
for _, tool_item in kept_tools:
|
||||
cid = tool_item.get("call_id", tool_item.get("id", ""))
|
||||
if cid not in added_call_ids:
|
||||
if cid not in added_pairs:
|
||||
result.append(tool_item)
|
||||
|
||||
for cs_item in compaction_summaries:
|
||||
@@ -5037,7 +5121,7 @@ def _antigravity_normalize_context(input_data, model=""):
|
||||
|
||||
while len(result) > _ANTIGRAVITY_MAX_CONTENTS and total_chars > _ANTIGRAVITY_SOFT_CHARS:
|
||||
for i in range(1, len(result) - 1):
|
||||
if isinstance(result[i], dict) and result[i].get("type") in ("message", "function_call_output"):
|
||||
if isinstance(result[i], dict) and result[i].get("type") in ("message",):
|
||||
removed = result.pop(i)
|
||||
total_chars -= len(json.dumps(removed, ensure_ascii=False))
|
||||
break
|
||||
@@ -5217,6 +5301,9 @@ class Handler(http.server.BaseHTTPRequestHandler):
|
||||
elif BACKEND in ("codebuff", "freebuff"):
|
||||
self._handle_codebuff(body, model, stream, tracker)
|
||||
elif (BACKEND or "").startswith("gemini-oauth"):
|
||||
if OAUTH_PROVIDER == "google-antigravity":
|
||||
self._handle_antigravity_v2(body, model, stream, tracker)
|
||||
else:
|
||||
self._handle_gemini_oauth(body, model, stream, tracker)
|
||||
else:
|
||||
self._handle_openai_compat(body, model, stream, tracker)
|
||||
@@ -5406,6 +5493,383 @@ class Handler(http.server.BaseHTTPRequestHandler):
|
||||
chat_body["reasoning_effort"] = REASONING_EFFORT
|
||||
return chat_body
|
||||
|
||||
def _handle_antigravity_v2(self, body, model, stream, tracker=None):
|
||||
input_data = body.get("input", "")
|
||||
_schema = _load_schema(model=model)
|
||||
if _schema and not _schema.supports_vision:
|
||||
input_data = _preprocess_vision_input(input_data, _schema)
|
||||
body = dict(body)
|
||||
body["input"] = input_data
|
||||
|
||||
if isinstance(input_data, list) and len(input_data) > 30:
|
||||
input_data = _antigravity_normalize_context(input_data, model)
|
||||
body = dict(body)
|
||||
body["input"] = input_data
|
||||
|
||||
access_token = _refresh_oauth_token()
|
||||
token_path = os.path.join(os.path.expanduser("~"), ".cache", "codex-proxy", "google-antigravity-oauth-token.json")
|
||||
project_id = ""
|
||||
try:
|
||||
with open(token_path) as f:
|
||||
project_id = json.load(f).get("project_id", "")
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
tool_call_names = {}
|
||||
contents = []
|
||||
|
||||
if isinstance(input_data, list):
|
||||
for item in input_data:
|
||||
t = item.get("type")
|
||||
if t == "message":
|
||||
role = "user" if item.get("role") == "user" else "model"
|
||||
content = item.get("content", "")
|
||||
parts = []
|
||||
if isinstance(content, list):
|
||||
for c in content:
|
||||
ct = c.get("type")
|
||||
if ct in ("input_text", "text"):
|
||||
parts.append({"text": c.get("text", "")})
|
||||
elif ct in ("input_image", "image_url"):
|
||||
iu = c.get("image_url") or c.get("url", {})
|
||||
url = iu.get("url", iu) if isinstance(iu, dict) else iu
|
||||
if isinstance(url, str) and url.startswith("data:"):
|
||||
mime, _, b64 = url.partition(";base64,")
|
||||
mime = mime.replace("data:", "") or "image/png"
|
||||
parts.append({"inlineData": {"mimeType": mime, "data": b64}})
|
||||
else:
|
||||
parts.append({"text": str(url)})
|
||||
elif isinstance(content, str):
|
||||
parts.append({"text": content})
|
||||
if parts:
|
||||
contents.append({"role": role, "parts": parts})
|
||||
elif t == "function_call":
|
||||
call_id = item.get("call_id") or item.get("id") or f"call_{uuid.uuid4().hex[:24]}"
|
||||
fname = item.get("name", "")
|
||||
if call_id and fname:
|
||||
tool_call_names[call_id] = fname
|
||||
args = item.get("arguments", "{}")
|
||||
if isinstance(args, str):
|
||||
try:
|
||||
args = json.loads(args)
|
||||
except Exception:
|
||||
args = {}
|
||||
fc_part = {"functionCall": {"name": fname, "args": args, "id": call_id}}
|
||||
stored_sig = _gemini_get_sig(f"fc:{call_id}") or _gemini_get_sig(f"fc:{fname}")
|
||||
if stored_sig:
|
||||
fc_part["thoughtSignature"] = stored_sig
|
||||
fc_part["thought_signature"] = stored_sig
|
||||
else:
|
||||
fc_part["thought_signature"] = "skip_thought_signature_validator"
|
||||
contents.append({"role": "model", "parts": [fc_part]})
|
||||
elif t == "function_call_output":
|
||||
call_id = item.get("call_id", item.get("id", ""))
|
||||
output = item.get("output", "")
|
||||
fname = item.get("name", "") or tool_call_names.get(call_id, "")
|
||||
resp_part = {"functionResponse": {"name": fname or "unknown", "response": {"result": str(output)}}}
|
||||
if call_id:
|
||||
resp_part["functionResponse"]["id"] = call_id
|
||||
contents.append({"role": "user", "parts": [resp_part]})
|
||||
|
||||
sanitized = []
|
||||
last_user_text = None
|
||||
last_role = None
|
||||
for content in contents:
|
||||
role = content.get("role")
|
||||
parts = [p for p in content.get("parts", []) if isinstance(p, dict)]
|
||||
if not parts:
|
||||
continue
|
||||
has_function_call = any("functionCall" in p for p in parts)
|
||||
has_function_response = any("functionResponse" in p for p in parts)
|
||||
text_key = "\n".join([p.get("text", "") for p in parts if "text" in p]).strip()
|
||||
|
||||
if has_function_call or has_function_response:
|
||||
sanitized.append({"role": role, "parts": parts})
|
||||
last_role = role
|
||||
continue
|
||||
|
||||
if role == "user" and text_key and text_key == last_user_text:
|
||||
continue
|
||||
|
||||
if role == last_role and role in ("user", "model") and sanitized:
|
||||
last_parts = sanitized[-1].get("parts", [])
|
||||
last_has_tool = any("functionCall" in p or "functionResponse" in p for p in last_parts)
|
||||
if not last_has_tool:
|
||||
sanitized[-1].setdefault("parts", []).extend(parts)
|
||||
if role == "user" and text_key:
|
||||
last_user_text = text_key
|
||||
last_role = role
|
||||
continue
|
||||
|
||||
sanitized.append({"role": role, "parts": parts})
|
||||
if role == "user" and text_key:
|
||||
last_user_text = text_key
|
||||
last_role = role
|
||||
|
||||
while sanitized and sanitized[0].get("role") != "user":
|
||||
sanitized.pop(0)
|
||||
|
||||
contents = sanitized
|
||||
|
||||
instructions = body.get("instructions", "").strip()
|
||||
ag_identity = "You are Antigravity, a powerful agentic AI coding assistant designed by the Google Deepmind team working on Advanced Agentic Coding.\nYou are pair programming with a USER to solve their coding task. The task may require creating a new codebase, modifying or debugging an existing codebase, or simply answering a question.\n**Absolute paths only**\n**Proactiveness**"
|
||||
system_parts = [{"text": ag_identity}, {"text": "\n--- [SYSTEM_PROMPT_END] ---"}]
|
||||
if instructions:
|
||||
system_parts.append({"text": instructions})
|
||||
|
||||
gen_config = {"maxOutputTokens": body.get("max_output_tokens", 64000), "stopSequences": ["\n\nHuman:", "[DONE]"]}
|
||||
if body.get("temperature") is not None:
|
||||
gen_config["temperature"] = body["temperature"]
|
||||
if body.get("top_p") is not None:
|
||||
gen_config["topP"] = body["top_p"]
|
||||
|
||||
_is_claude_model = "claude" in model.lower()
|
||||
_is_claude_thinking = _is_claude_model and "thinking" in model.lower()
|
||||
|
||||
if REASONING_ENABLED and REASONING_EFFORT != "none":
|
||||
if _is_claude_thinking:
|
||||
budget = {"low": 8192, "medium": 16384, "high": 32768}.get(REASONING_EFFORT, 16384)
|
||||
gen_config["thinkingConfig"] = {"include_thoughts": True, "thinking_budget": budget}
|
||||
if gen_config.get("maxOutputTokens", 0) <= budget:
|
||||
gen_config["maxOutputTokens"] = 64000
|
||||
elif not _is_claude_model:
|
||||
budget = {"low": 2048, "medium": 8192, "high": 24576}.get(REASONING_EFFORT, 8192)
|
||||
gen_config["thinkingConfig"] = {"includeThoughts": True, "thinkingBudget": budget}
|
||||
|
||||
oa_tools = body.get("tools", [])
|
||||
gemini_tools = []
|
||||
if oa_tools:
|
||||
func_decls = []
|
||||
for tool in oa_tools:
|
||||
ttype = tool.get("type", "function")
|
||||
fname = tool.get("name", "")
|
||||
if ttype == "function":
|
||||
fn = tool.get("function", tool)
|
||||
name = fn.get("name", fname)
|
||||
desc = fn.get("description", "")
|
||||
params = fn.get("parameters", fn.get("input_schema", {}))
|
||||
func_decls.append({"name": name, "description": desc, "parameters": params})
|
||||
elif fname:
|
||||
func_decls.append({"name": fname, "description": tool.get("description", ""), "parameters": tool.get("parameters", {"type": "object", "properties": {}})})
|
||||
if func_decls:
|
||||
gemini_tools = [{"functionDeclarations": func_decls}]
|
||||
|
||||
contents = _gemini_reattach_sigs(contents)
|
||||
|
||||
ag_key = _antigravity_loop_key(self._session_id)
|
||||
with _ANTIGRAVITY_LOOP_TRACKER_LOCK:
|
||||
if ag_key not in _ANTIGRAVITY_LOOP_TRACKER:
|
||||
_ANTIGRAVITY_LOOP_TRACKER[ag_key] = {
|
||||
"latest_user_hash": None, "nudge_injected": False, "latest_user_appended": False,
|
||||
"tool_calls_for_request": 0, "repeated_tool": False, "force_finalize": False,
|
||||
"last_tool": None, "last_tool_count": 0,
|
||||
}
|
||||
ag_state = _ANTIGRAVITY_LOOP_TRACKER[ag_key]
|
||||
|
||||
latest_user = ""
|
||||
if isinstance(input_data, list):
|
||||
for item in reversed(input_data):
|
||||
if item.get("type") == "message" and item.get("role") == "user":
|
||||
c = item.get("content", "")
|
||||
if isinstance(c, str):
|
||||
latest_user = c
|
||||
elif isinstance(c, list):
|
||||
latest_user = "\n".join(p.get("text", p.get("input_text", "")) for p in c if isinstance(p, dict))
|
||||
break
|
||||
if latest_user:
|
||||
latest_norm = " ".join(latest_user.strip().split())[:200]
|
||||
latest_user_hash = hashlib.sha256(latest_norm.encode()).hexdigest()[:16]
|
||||
if latest_user_hash != ag_state.get("latest_user_hash"):
|
||||
ag_state["latest_user_hash"] = latest_user_hash
|
||||
ag_state["nudge_injected"] = False
|
||||
ag_state["latest_user_appended"] = False
|
||||
ag_state["tool_calls_for_request"] = 0
|
||||
ag_state["repeated_tool"] = False
|
||||
ag_state["force_finalize"] = False
|
||||
ag_state["last_tool"] = None
|
||||
ag_state["last_tool_count"] = 0
|
||||
n_tool_calls = sum(1 for it in input_data if isinstance(it, dict) and it.get("type") == "function_call")
|
||||
ag_state["tool_calls_for_request"] = n_tool_calls
|
||||
last_tool_key = None
|
||||
for item in reversed(input_data):
|
||||
if isinstance(item, dict) and item.get("type") == "function_call":
|
||||
fname = item.get("name", "")
|
||||
args_str = json.dumps(item.get("arguments", {}), sort_keys=True)[:100]
|
||||
last_tool_key = f"{fname}:{args_str}"
|
||||
break
|
||||
if last_tool_key:
|
||||
if last_tool_key == ag_state.get("last_tool"):
|
||||
ag_state["last_tool_count"] = ag_state.get("last_tool_count", 0) + 1
|
||||
if ag_state["last_tool_count"] >= 5:
|
||||
ag_state["repeated_tool"] = True
|
||||
ag_state["force_finalize"] = True
|
||||
else:
|
||||
ag_state["last_tool"] = last_tool_key
|
||||
ag_state["last_tool_count"] = 1
|
||||
|
||||
if ag_state.get("force_finalize"):
|
||||
contents.append({"role": "user", "parts": [{"text": "STOP CALLING TOOLS. APPLY THE FINAL EDIT OR SUMMARIZE WHAT BLOCKED YOU. DO NOT CALL ANY MORE TOOLS."}]})
|
||||
|
||||
if not _antigravity_is_simple_user(latest_user):
|
||||
contents.insert(0, {"role": "user", "parts": [{"text": _GEMINI_AGENT_GUARDRAIL}]})
|
||||
|
||||
request_body = {"contents": contents, "safetySettings": [
|
||||
{"category": "HARM_CATEGORY_HARASSMENT", "threshold": "OFF"},
|
||||
{"category": "HARM_CATEGORY_HATE_SPEECH", "threshold": "OFF"},
|
||||
{"category": "HARM_CATEGORY_SEXUALLY_EXPLICIT", "threshold": "OFF"},
|
||||
{"category": "HARM_CATEGORY_DANGEROUS_CONTENT", "threshold": "OFF"},
|
||||
{"category": "HARM_CATEGORY_CIVIC_INTEGRITY", "threshold": "OFF"},
|
||||
]}
|
||||
request_body["systemInstruction"] = {"role": "user", "parts": system_parts}
|
||||
if gen_config:
|
||||
request_body["generationConfig"] = gen_config
|
||||
if gemini_tools:
|
||||
request_body["tools"] = gemini_tools
|
||||
if _is_claude_model and gemini_tools:
|
||||
request_body["toolConfig"] = {"functionCallingConfig": {"mode": "VALIDATED"}}
|
||||
|
||||
import platform as _plat
|
||||
_os_name = _plat.system().lower()
|
||||
_os_arch = _plat.machine().lower().replace("x86_64", "x64").replace("aarch64", "arm64")
|
||||
_fetched_ver = _ensure_antigravity_version()
|
||||
_ag_ua = f"antigravity/{_fetched_ver} {_os_name}/{_os_arch}"
|
||||
headers = {
|
||||
"Content-Type": "application/json",
|
||||
"Authorization": f"Bearer {access_token}",
|
||||
"User-Agent": _ag_ua,
|
||||
"X-Client-Name": "antigravity",
|
||||
"X-Client-Version": _ensure_antigravity_client_version(),
|
||||
"x-goog-api-client": "gl-node/18.18.2 fire/0.8.6 grpc/1.10.x",
|
||||
}
|
||||
|
||||
wrapped = {
|
||||
"project": project_id,
|
||||
"model": model,
|
||||
"requestType": "agent",
|
||||
"userAgent": _ag_ua,
|
||||
"requestId": f"agent-{uuid.uuid4().hex[:12]}",
|
||||
"request": request_body,
|
||||
}
|
||||
wrapped["request"]["sessionId"] = f"{uuid.uuid4().hex}{int(time.time()*1000)}"
|
||||
|
||||
_allow_staging = os.environ.get("ALLOW_ANTIGRAVITY_STAGING", "0") == "1"
|
||||
_antigravity_endpoints = [
|
||||
"https://daily-cloudcode-pa.googleapis.com",
|
||||
"https://daily-cloudcode-pa.sandbox.googleapis.com",
|
||||
"https://cloudcode-pa.googleapis.com",
|
||||
]
|
||||
if _allow_staging:
|
||||
_antigravity_endpoints.append("https://autopush-cloudcode-pa.sandbox.googleapis.com")
|
||||
|
||||
body_b = json.dumps(wrapped).encode()
|
||||
print(f"[{self._session_id}] [antigravity-v2] model={model} stream={stream} contents={len(contents)} tools={bool(gemini_tools)} project={project_id} ver={_versions[0]}", file=sys.stderr)
|
||||
try:
|
||||
debug_path = os.path.join(_LOG_DIR, f"antigravity-v2-request-{self._session_id}.json")
|
||||
with open(debug_path, "w") as dbg:
|
||||
json.dump(wrapped, dbg, indent=2)
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
upstream = None
|
||||
chosen_ep = None
|
||||
global _antigravity_preferred_endpoint
|
||||
with _antigravity_endpoint_lock:
|
||||
_pref = _antigravity_preferred_endpoint
|
||||
ordered = ([_pref] + [e for e in _antigravity_endpoints if e != _pref]) if _pref and _pref in _antigravity_endpoints else list(_antigravity_endpoints)
|
||||
|
||||
_all_404 = True
|
||||
for ep in ordered:
|
||||
action = "streamGenerateContent" if stream else "generateContent"
|
||||
url_suffix = f"v1internal:{action}?alt=sse" if stream else f"v1internal:{action}"
|
||||
target = f"{ep}/{url_suffix}"
|
||||
req = urllib.request.Request(target, data=body_b, headers=headers)
|
||||
try:
|
||||
upstream = urllib.request.urlopen(req, timeout=_upstream_timeout(body, stream))
|
||||
chosen_ep = ep
|
||||
_all_404 = False
|
||||
with _antigravity_endpoint_lock:
|
||||
_antigravity_preferred_endpoint = ep
|
||||
break
|
||||
except urllib.error.HTTPError as e:
|
||||
err_body = e.read().decode()
|
||||
err_class = _classify_antigravity_error(e.code, err_body)
|
||||
print(f"[{self._session_id}] [antigravity-v2] {ep.replace('https://','')} {e.code} class={err_class} body={err_body[:300]}", file=sys.stderr)
|
||||
if e.code != 404:
|
||||
_all_404 = False
|
||||
if e.code in (400, 404):
|
||||
try:
|
||||
debug_path = os.path.join(_LOG_DIR, f"antigravity-v2-{e.code}.json")
|
||||
with open(debug_path, "w") as dbg:
|
||||
json.dump({"endpoint": ep, "url": target, "model": model, "wrapped": wrapped, "error": err_body}, dbg, indent=2)
|
||||
except Exception:
|
||||
pass
|
||||
if e.code == 400:
|
||||
return self.send_json(e.code, {"error": {"type": "upstream_error", "message": _sanitize_err_body(err_body)}})
|
||||
if err_class in ("auth_permanent", "service_disabled", "forbidden", "account_banned", "validation_required"):
|
||||
return self.send_json(e.code, {"error": {"type": "upstream_error", "message": _sanitize_err_body(err_body)}})
|
||||
if err_class in ("quota_exhausted", "rate_limited"):
|
||||
pool = _google_antigravity_pool
|
||||
_, acct = _get_google_account(OAUTH_PROVIDER)
|
||||
if acct:
|
||||
reset_s = _parse_rate_limit_reset(err_body)
|
||||
cooldown = reset_s if reset_s and reset_s > 10 else 60
|
||||
pool.mark_rate_limited(acct, cooldown)
|
||||
return self.send_json(e.code, {"error": {"type": "upstream_error", "message": _sanitize_err_body(err_body)}})
|
||||
if ep == ordered[-1] and not _all_404:
|
||||
return self.send_json(e.code, {"error": {"type": "upstream_error", "message": _sanitize_err_body(err_body)}})
|
||||
continue
|
||||
except Exception as e:
|
||||
_all_404 = False
|
||||
print(f"[{self._session_id}] [antigravity-v2] {ep.replace('https://','')} conn failed: {e}", file=sys.stderr)
|
||||
if ep == ordered[-1]:
|
||||
return self.send_json(502, {"error": {"type": "proxy_error", "message": str(e)}})
|
||||
continue
|
||||
|
||||
if _all_404 and upstream is None:
|
||||
print(f"[{self._session_id}] [antigravity-v2] all endpoints 404, invalidating version cache and re-fetching", file=sys.stderr)
|
||||
global _antigravity_version_validated
|
||||
with _antigravity_version_lock:
|
||||
_antigravity_version_validated = False
|
||||
_antigravity_version_checked = 0
|
||||
_new_ver = _ensure_antigravity_version()
|
||||
if _new_ver != _fetched_ver:
|
||||
print(f"[{self._session_id}] [antigravity-v2] version changed {_fetched_ver} -> {_new_ver}, retrying", file=sys.stderr)
|
||||
_ag_ua_new = f"antigravity/{_new_ver} {_os_name}/{_os_arch}"
|
||||
headers["User-Agent"] = _ag_ua_new
|
||||
wrapped["userAgent"] = _ag_ua_new
|
||||
body_b = json.dumps(wrapped).encode()
|
||||
for ep in ordered:
|
||||
action = "streamGenerateContent" if stream else "generateContent"
|
||||
url_suffix = f"v1internal:{action}?alt=sse" if stream else f"v1internal:{action}"
|
||||
target = f"{ep}/{url_suffix}"
|
||||
req = urllib.request.Request(target, data=body_b, headers=headers)
|
||||
try:
|
||||
upstream = urllib.request.urlopen(req, timeout=_upstream_timeout(body, stream))
|
||||
chosen_ep = ep
|
||||
with _antigravity_endpoint_lock:
|
||||
_antigravity_preferred_endpoint = ep
|
||||
break
|
||||
except urllib.error.HTTPError as e:
|
||||
err_body = e.read().decode()
|
||||
print(f"[{self._session_id}] [antigravity-v2-retry] {ep.replace('https://','')} {e.code}", file=sys.stderr)
|
||||
if e.code == 400:
|
||||
return self.send_json(e.code, {"error": {"type": "upstream_error", "message": _sanitize_err_body(err_body)}})
|
||||
if ep == ordered[-1]:
|
||||
return self.send_json(e.code, {"error": {"type": "upstream_error", "message": _sanitize_err_body(err_body)}})
|
||||
continue
|
||||
except Exception as e:
|
||||
if ep == ordered[-1]:
|
||||
return self.send_json(502, {"error": {"type": "proxy_error", "message": str(e)}})
|
||||
continue
|
||||
|
||||
if upstream is None:
|
||||
return self.send_json(502, {"error": {"type": "proxy_error", "message": "All endpoints failed"}})
|
||||
|
||||
if stream:
|
||||
self._forward_gemini_sse(upstream, model, body, input_data, tracker)
|
||||
else:
|
||||
self._forward_gemini_json(upstream, model, body, input_data)
|
||||
|
||||
def _handle_gemini_oauth(self, body, model, stream, tracker=None):
|
||||
input_data = body.get("input", "")
|
||||
policy = provider_policy()
|
||||
@@ -5573,6 +6037,9 @@ class Handler(http.server.BaseHTTPRequestHandler):
|
||||
resp_part["functionResponse"]["id"] = call_id
|
||||
contents.append({"role": "user", "parts": [resp_part]})
|
||||
|
||||
# CRITICAL FIX: Sanitize contents while PRESERVING functionCall -> functionResponse alternation.
|
||||
# Google's Gemini API REQUIRES: functionCall (role=model) must be immediately followed by functionResponse (role=user).
|
||||
# We NEVER merge, skip, or reorder tool-related messages.
|
||||
if OAUTH_PROVIDER.startswith("google") and "claude" not in model.lower():
|
||||
sanitized = []
|
||||
last_user_text = None
|
||||
@@ -5582,24 +6049,36 @@ class Handler(http.server.BaseHTTPRequestHandler):
|
||||
parts = [p for p in content.get("parts", []) if isinstance(p, dict)]
|
||||
if not parts:
|
||||
continue
|
||||
# Check if this content has functionCall or functionResponse - these must be preserved
|
||||
# Check if this content has functionCall or functionResponse - these MUST be preserved as-is
|
||||
has_function_call = any("functionCall" in p for p in parts)
|
||||
has_function_response = any("functionResponse" in p for p in parts)
|
||||
text_key = "\n".join([p.get("text", "") for p in parts if "text" in p]).strip()
|
||||
# Skip duplicate user text messages, but NEVER skip function responses
|
||||
if role == "user" and text_key and text_key == last_user_text and not has_function_response:
|
||||
|
||||
# Tool calls/responses are NEVER merged or skipped - they must maintain strict order
|
||||
if has_function_call or has_function_response:
|
||||
sanitized.append({"role": role, "parts": parts})
|
||||
continue
|
||||
# CRITICAL FIX: Function calls (model role) and function responses (user role) MUST NOT be merged
|
||||
# Google's API requires strict alternation: functionCall (model) -> functionResponse (user)
|
||||
# Never merge across role boundaries when function calls/responses are involved
|
||||
if role == last_role and role in ("user", "model") and sanitized and not has_function_call and not has_function_response:
|
||||
# Only merge same-role consecutive text messages without tool content
|
||||
|
||||
# For plain text messages only: skip duplicate consecutive user text
|
||||
if role == "user" and text_key and text_key == last_user_text:
|
||||
continue
|
||||
|
||||
# Merge consecutive same-role TEXT-ONLY messages (no tool content)
|
||||
if role == last_role and role in ("user", "model") and sanitized:
|
||||
last_parts = sanitized[-1].get("parts", [])
|
||||
# Only merge if the last message is also text-only (no functionCall/functionResponse)
|
||||
last_has_tool = any("functionCall" in p or "functionResponse" in p for p in last_parts)
|
||||
if not last_has_tool:
|
||||
sanitized[-1].setdefault("parts", []).extend(parts)
|
||||
else:
|
||||
if role == "user" and text_key:
|
||||
last_user_text = text_key
|
||||
continue
|
||||
|
||||
sanitized.append({"role": role, "parts": parts})
|
||||
if role == "user" and text_key:
|
||||
last_user_text = text_key
|
||||
last_role = role
|
||||
|
||||
# Trim leading non-user messages (Google expects conversation to start with user)
|
||||
while sanitized and sanitized[0].get("role") != "user":
|
||||
sanitized.pop(0)
|
||||
|
||||
Reference in New Issue
Block a user