Skip to content

armgit / avibe

Public Imported from GitHub
0

Make your computer a home for your AI partner. Avibe is the local-first Agent OS: Vibey, its built-in AI partner, works out of the box, and Claude Code, Codex & OpenCode run unchanged beside it — reach them from your browser, phone, or any chat app.

test(agent-core): reconcile prefix exclusions with request-code priority (#3079)

Call White2e86c967e6f1
diff --git a/tests/agent_core/agent/test_compaction.py b/tests/agent_core/agent/test_compaction.py
index b14457336..c2c907d7d 100644
--- a/tests/agent_core/agent/test_compaction.py
+++ b/tests/agent_core/agent/test_compaction.py
@@ -953,15 +953,26 @@ async def test_two_ineffective_checkpoints_stop_compaction_for_the_run():
@pytest.mark.parametrize(
- ("prefix", "ending", "attempts"),
- [("Throttling error", "error", 3), ("Service unavailable", "completed", 4)],
+ ("prefix", "code", "ending", "attempts"),
+ [
+ ("Throttling error", None, "error", 3),
+ ("Service unavailable", None, "completed", 4),
+ ("Throttling error", "invalid_request_error", "error", 3),
+ ("Service unavailable", "invalid_request_error", "error", 3),
+ ("Throttling error", "upstream_request_invalid", "error", 3),
+ ("Service unavailable", "upstream_request_invalid", "error", 3),
+ ],
)
-async def test_json_non_overflow_errors_use_ordinary_handling_without_context_relief(prefix, ending, attempts):
+async def test_json_non_overflow_errors_use_ordinary_handling_without_context_relief(prefix, code, ending, attempts):
# Enough context for the overflow ladder to checkpoint, but below the proactive threshold. Feed the real
# classifier's outcome into the loop, not a hand-written ProviderError that assumes classification is correct.
+ # Only the untyped service message uses the transient heuristic; explicit request codes forbid replay.
+ error = {"message": f"{prefix}: Too many tokens, please wait before trying again."}
+ if code is not None:
+ error["type"] = code
refusal = classify_error(
status=400,
- body=json.dumps(anthropic_error(f"{prefix}: Too many tokens, please wait before trying again.")),
+ body=json.dumps({"type": "error", "error": error}),
protocol="anthropic",
)
model = Model(
diff --git a/tests/agent_core/ai/test_transform_sse_errors.py b/tests/agent_core/ai/test_transform_sse_errors.py
index 1d46f6e86..6b9221cd9 100644
--- a/tests/agent_core/ai/test_transform_sse_errors.py
+++ b/tests/agent_core/ai/test_transform_sse_errors.py
@@ -358,7 +358,7 @@ def test_overflow_classifies_cerebras_bodyless_status() -> None:
@pytest.mark.parametrize("status", [400, 429])
[email protected]("representation", ["plain", "json", "json-string"])
[email protected]("representation", ["plain", "json", "json-string", "json-invalid-request"])
@pytest.mark.parametrize(
("message", "kind", "retryable"),
[
@@ -375,11 +375,17 @@ def test_non_overflow_prefixes_survive_message_extraction(
) -> None:
body = {
"plain": message,
- "json": json.dumps({"type": "error", "error": {"type": "invalid_request_error", "message": message}}),
+ "json": json.dumps({"type": "error", "error": {"message": message}}),
"json-string": json.dumps(message),
+ "json-invalid-request": json.dumps(
+ {"type": "error", "error": {"type": "invalid_request_error", "message": message}}
+ ),
}[representation]
if status == 429:
kind, retryable = "rate_limit", True
+ elif representation == "json-invalid-request":
+ # Excluding overflow does not make a typed request rejection transient, even if its text says unavailable.
+ kind, retryable = "invalid_request", False
error = classify_error(status=status, body=body)
@@ -419,6 +425,16 @@ def test_non_message_json_fields_cannot_supply_a_non_overflow_prefix(prefix: str
("status", "code", "message", "kind"),
[
(400, "ThrottlingException", "too many tokens", "invalid_request"),
+ (None, "invalid_request_error", "Service unavailable: too many tokens", "invalid_request"),
+ (None, "invalid_request_error", "Throttling error: too many tokens", "invalid_request"),
+ (None, "upstream_request_invalid", "Service unavailable: too many tokens", "invalid_request"),
+ (None, "upstream_request_invalid", "Throttling error: too many tokens", "invalid_request"),
+ (400, "upstream_request_invalid", "Service unavailable: too many tokens", "invalid_request"),
+ (400, "upstream_request_invalid", "Throttling error: too many tokens", "invalid_request"),
+ (None, "invalid_request_error", "too many tokens", "overflow"),
+ (400, "invalid_request_error", "too many tokens", "overflow"),
+ (None, "upstream_request_invalid", "too many tokens", "overflow"),
+ (400, "upstream_request_invalid", "too many tokens", "overflow"),
(400, "context_length_exceeded", "Throttling error: too many tokens", "overflow"),
(400, "request_too_large", "Service unavailable: too many tokens", "overflow"),
(400, "rate_limit_error", "Throttling error: too many tokens", "rate_limit"),
@@ -432,10 +448,13 @@ def test_non_message_json_fields_cannot_supply_a_non_overflow_prefix(prefix: str
],
)
def test_non_overflow_prefixes_preserve_status_and_machine_code_priority(
- status: int, code: str, message: str, kind: str,
+ status: int | None, code: str, message: str, kind: str,
) -> None:
body = json.dumps({"error": {"type": code, "message": message}})
- assert classify_error(status=status, body=body).kind == kind
+ error = classify_error(status=status, body=body)
+ assert error.kind == kind
+ assert error.retryable is (kind in {"rate_limit", "server"})
+ assert error.continuable is (kind in {"rate_limit", "server"})
@pytest.mark.parametrize("status", [401, 403])