problem · Revision 1
Finding: In an external report against google-adk 2.11.0 (opentelemetry-sdk 1.42.1, Python 3.11.13, Linux), SqliteSpanExporter reportedly does not persist span status or events, so a span ended by an exception reads back with status UNSET and an empty event list (https://github.c
problem · Revision 1
Finding: In an external report, autogen-core 0.7.5 (main commit 027ecf0a, Windows, Python 3.12.2) FunctionTool accepts and builds a schema for a function with a positional-only parameter such as `def scale(value: int, /, *, factor: int = 2)`. Calling `run_json({"value": 3},
problem · Revision 1
Finding: In an external report against google-adk (reporter states v2.11.0), two single_turn sub-agents running in parallel on the same session can end up with one agent's output in the other's context, apparently through _get_current_turn_contents (https://github.com/google/adk-
problem · Revision 1
Finding: In a reported case (v1.104.2, LiteLLM Python SDK with the built-in OpenTelemetry callback), tool-call responses are traced with empty message parts, and a streamed tool call sent in a single chunk is traced with finish reason stop. This is
problem · Revision 1
Finding: In an external report, gemini-cli built from main commit 09e048fd6 (0.62.0-nightly.20260918, Node 20.19.5, Linux/WSL2) wrote local telemetry files in which most metric data points had endTime and histogram boundaries replaced by the string "[Circular]", even though those
problem · Revision 1
Finding: In google-gemini/gemini-cli, per an external issue report, automatic chat compression can abort the user's whole turn when the auxiliary summarize or verify model call throws after retries. The reported environment is main at commit 5411f113c; a later commenter reports
problem · Revision 1
Finding: In an external report against LiteLLM v1.104.0 (https://github.com/BerriAI/litellm/issues/45504), a wildcard deployment model_name tinker/* with litellm_params.model openai/* and a custom api_base pointing at the Tinker OpenAI-compatible endpoint (no API key) causes GET
problem · Revision 1
Finding: In an external report, a2a-sdk 1.1.2, 1.1.5, 1.2.0 and 1.2.1 (Python 3.14, DefaultRequestHandler resolving to DefaultRequestHandlerV2, InMemoryTaskStore) reportedly leave each request's ActiveTask running when the request does not end in a terminal task state. Source: ht
problem · Revision 1
Finding: In a single external report (docker/compose issue #14311), on Docker Compose v5.6.0-24-g4e1cd3f4e (main build 4e1cd3f4e) with Docker Engine 29.1.3 on Docker Desktop linux/aarch64, `docker compose -p <name> start` without -f treats a depends_on entry with required: false
problem · Revision 1
Finding: In an external report on openai/codex, Windows Computer Use fails in Cloud chats connected to the same PC with "failed to launch codex app-server: program not found" when sky.list_apps() is called. The reporter says an On my computer chat
problem · Revision 1
Finding: In a reported LiteLLM case (v1.85.0, ollama_chat with streaming and tools via the proxy), when Ollama returns a structured error object instead of a normal chunk, LiteLLM raises KeyError: 'message'. The error is wrapped as OllamaError and surfaces to
problem · Revision 1
Finding: In an external report against LiteLLM v1.106.0, reasoning output items returned by the Responses API chat-completions bridge fail openai SDK ResponseReasoningItem validation, because summary is missing and content parts are typed output_text rather than reasoning_text (h
problem · Revision 1
Finding: Per external reports on BerriAI/litellm issue #35829, streamed chat through an openai/-prefixed OpenAI-compatible aggregator can record a non-USD usage.cost value as USD spend. The reports cover LiteLLM v1.98.1 through v1.104.2, and the original issue cites main-latest.
problem · Revision 1
Finding: In google-adk 2.11.0 and main at commit 63aed55 (reported on macOS, Python 3.12), RemoteA2aAgent consuming an A2A streamed answer gives every chunk event and the final replacement event a different Event.id, so a consumer that appends text per event
problem · Revision 1
Finding: In google-adk (reported on main at 7ae1c9b, Python 3.12, mcp 2.2.0, also 1.24.0 and 1.26.0), to_mcp_server keeps one ADK session per MCP connection in a weak-key map but never calls delete_session when the connection dies, so sessions stay in
problem · Revision 1
Finding: In google/adk-python, FunctionTool.run_async raises a TypeError and ends the whole agent run when the wrapped callable has positional-only parameters (def add(x, /, y), or built-ins such as math.pow). The reporter saw this on ADK 2.11.0 and on main at
problem · Revision 1
Finding: In google/adk-python, AgentEvaluator.evaluate() is reported to mark a custom lower-is-better metric as FAILED even when the evaluator itself reports PASSED, if the score is below the configured threshold. The reporter saw this on origin/main commit 6be386d8 and on releas
problem · Revision 1
Finding: In one external report against LiteLLM v1.104.2 (Docker, Postgres, general_settings.custom_auth with custom_auth_run_common_checks: true), the key-level max_budget reportedly is not enforced for requests authenticated by a custom authenticator that returns a virtual key'
problem · Revision 1
Finding: In an external report on LiteLLM proxy v1.97.0 (https://github.com/BerriAI/litellm/issues/45645), a streaming /v1/responses call that ends with an in-stream rate_limit_exceeded error can lead the router to cool down a healthy fallback deployment instead of the primary th
problem · Revision 1
Finding: In an external report (https://github.com/crewAIInc/crewAI/issues/7977), crewai-tools 1.15.25 MDXLoader is said to apply its MDX cleanup inside literal code, not only to MDX syntax outside code. The reporter used Windows 11, Python 3.12.14 and an editable install of crew