problem · Revision 2
Finding: According to the official Claude Code v2.1.290 release note (https://github.com/anthropics/claude-code/releases/tag/v2.1.290), headless runs using --json-schema exited non-zero with is_error: true on a success result when the connection dropped after the structured outpu
problem · Revision 1
Finding: The official release notes for @anthropic-ai/claude-agent-sdk v0.3.287 (published 2026-10-01) say the release fixed a tool call to an in-process MCP server being left waiting after toggleMcpServer() disabled that server or setMcpServers() removed it. The earliest affecte
problem · Revision 1
Finding: According to the official Claude Code v2.1.292 release note (https://github.com/anthropics/claude-code/releases/tag/v2.1.292), an MCP tool whose name was longer than 128 characters made every request fail in earlier versions. The note says v2.1.292 now leaves that tool o
problem · Revision 2
Finding: In an external report, Claude Code's Grep tool (version 2.1.295, Claude Desktop Code tab, Windows 11) returns "No matches found" for a pattern ending in $ (APPLE$) on a file with CRLF line endings, while the unanchored pattern APPLE
problem · Revision 2
Finding: In an external report against openai/openai-agents-js, the aisdk() adapter from @openai/agents-extensions (reported on 0.17.2, with the same code said to remain in 0.19.0) converts each function_call_result item into its own role 'tool' message. With Gemini 3.x through @
problem · Revision 1
Finding: In an external report against llama-index-core 0.14.25, RecursiveRetriever reportedly returns a resolved child node with score 1.0 when the parent IndexNode's NodeWithScore score is 0.0, instead of preserving 0.0 (https://github.com/run-llama/llama_index/issues/23429). T
problem · Revision 2
Finding: In openai-agents-python 0.23.1 (the reporter also says it reproduced on main at commit 125efa029b4bfd84238bd2c4fd69c3406f802663), the reporter says the Chat Completions converter Converter.items_to_messages produces an invalid message order when a Responses-format histor
problem · Revision 1
Finding: In an external report on openai/codex issue 52407, a Windows 25H2 (build 26200.9457) Codex Desktop 26.1002.7124.0 install with bundled core 0.162.0-alpha.2 fails its first Computer Use call, cua.getState(), with an MXC launcher error (HRESULT 0x80070003) even though a di
problem · Revision 2
Finding: In Playwright Test 1.64.0 on Node 22.22.0 and Ubuntu 24.04 (as reported), running npx playwright test --reporter=json from the project root with testDir set to ./tests and two projects (chromium, firefox) reportedly yields one spec per project instead of
problem · Revision 1
Finding: In an external report on @modelcontextprotocol/client 2.3.1 (Node 24.18.0, Linux x64, two CPUs, default Ajv), the first ordinary callTool after complete discovery reportedly prepares output validators for every cached tool, including tools never called, causing substanti
problem · Revision 2
Finding: In one external report on uv 0.12.23 (x86_64-unknown-linux-musl, Docker debian:trixie on GKE-based CI runners, Python 3.11), uv sync with UV_GCS_ENDPOINT_URL pointed at Google Artifact Registry and GOOGLE_APPLICATION_CREDENTIALS set to a workload identity federation (ext
problem · Revision 1
Finding: In an external report against uv 0.12.23 (x86_64 Linux), `uv build --all-packages` on a 54-member workspace (all hatchling plus hatch-vcs) intermittently hangs forever when the cache is cold (UV_NO_CACHE=1 or a fresh UV_CACHE_DIR); the cause is a reporter hypothesis,
problem · Revision 1
Finding: In an external report on the Claude desktop app Code tab (macOS, bundled Claude Code 2.1.295), a mod's ui.render hook on the AbovePrompt component reportedly draws its band only in the left pane of split view. The right pane
problem · Revision 1
Finding: One reporter says that after updating to Claude Code 2.1.296 on 2026-10-10 (macOS, terminal, claude.ai login, model claude-opus-5-5), auto-compaction fired well below the documented default of roughly 967K tokens for native 1M-window models. This is an external report an
problem · Revision 1
Finding: In the MCP Python SDK 2.1.1 (reported with Python 3.11.14, FastMCP 4.0.3, Streamable HTTP), a notification POST such as notifications/initialized can get HTTP 202 and then fail if the internal session writer is already closed, producing a duplicate-response error
problem · Revision 1
Finding: In the MCP Python SDK, an MCPServer @mcp.completion() handler that returns more than 100 values is reported to behave inconsistently by protocol version. On a 2026-07-28 session the whole completion/complete request fails with a generic error, and on a
problem · Revision 2
Finding: One external report says that on the mcp Python SDK SSE transport (FastMCP/sse_app, an older mcp 1.x release pinned via mcp<2, Windows 11, Python 3.14, uvicorn), a single request to /sse with a Host header rejected by DNS-rebinding protection
problem · Revision 1
Finding: On modelcontextprotocol/python-sdk v2 main at commit 91941ed (Python 3.12, macOS arm64, per the reporter), the Streamable HTTP server transport reportedly answers a POST with a mixed-case Content-Type such as Application/JSON with HTTP 415, so the initialize handshake fa
problem · Revision 1
Finding: According to an external report (not reproduced by KFA), the Python MCP SDK's MCPServer returns the wrong JSON-RPC error code for prompts/get when the prompt name is unregistered or a required argument is missing. The reporter observed this on
problem · Revision 1
Finding: In a single fresh external report (GitHub issue #101069, opened 2026-10-10), the Claude Code desktop app on a headless Ubuntu server with no desktop environment fails Google sign-in at the browser-to-app handoff. After the user completes Google consent in