Trending

Fixes people marked as working

Ranked by how many people marked that the fix worked in the last 7 days, then by how recently the status was checked. A chip is shown only for a repository that has a guide.

• All 107
• LangGraph 13
• MCP 12
• Pydantic AI 12
• Browser Use 10
• Hermes 9
• MCP Python SDK 9
• AutoGen 8
• CrewAI 7
• LangChain 7
• Coding agent CLI 6
• n8n 6
• OpenClaw 4
• Aider 2
• Continue 2

Ranked guides

• LangGraphFixed (verified)

  'This stream has already been locked' when streaming a JS graph in React Native

  Streaming a LangGraph.js graph in React Native fails with 'This stream has already been locked for exclusive reading by another reader'.

  Last checked Oct 17, 2026#transport#setup1 steps →0
• LangGraphFixed (verified)

  @langchain/langgraph 1.0.12 installs without its dist folder

  After installing @langchain/langgraph 1.0.12, imports fail because the package has no dist folder.

  Last checked Oct 17, 2026#setup#regression2 steps →0
• AiderFixed (verified)

  /read-only says 'No matches found' for paths containing [brackets]

  Adding a file with square brackets in its path (for example a [id] route folder) via /read-only fails with 'No matches found', while /add works.

  Last checked Oct 17, 2026#config1 steps →0
• LangGraphFixed (verified)

  AbortSignal passed to a JS graph doesn't stop it

  Passing an AbortSignal in config to a LangGraph.js graph doesn't cancel the run as expected.

  Last checked Oct 17, 2026#config#loop2 steps →0
• LangChainFixed (verified)

  After a human edits a tool call, the agent runs the original call anyway

  With HumanInTheLoopMiddleware, choosing 'edit' runs the edited tool call, but on the next step the agent issues the original, unedited call again.

  Last checked Oct 17, 2026#tool-calls#state1 steps →0
• Pydantic AIFixed (verified)

  AG-UI tool calls point to a parent message that was never announced

  For a response with only tool calls, the AG-UI event stream emits TOOL_CALL_START with a parentMessageId that no TEXT_MESSAGE_START ever announced.

  Last checked Oct 17, 2026#messaging#state1 steps →0
• n8nFixed (verified)

  AI Agent with a structured output parser fails after a few turns with nested 'output'

  An AI Agent using a structured output parser plus memory works at first, then starts failing schema validation after several turns.

  Last checked Oct 17, 2026#parsing#state2 steps →0
• n8nFixed (verified)

  AI nodes fail after 5 minutes with UND_ERR_HEADERS_TIMEOUT

  Any AI node calling a slow model (often a local one) fails once a response takes longer than five minutes.

  Last checked Oct 17, 2026#config#transport2 steps →0
• OpenClawFixed (verified)

  All CLI commands fail with 'pairing required' after updating the gateway

  After updating to 2026.2.19-2, CLI commands fail with 'gateway closed (1008): pairing required' even though the gateway is running.

  Last checked Oct 17, 2026#regression#config1 steps →0
• browser-useFixed (verified)

  Browser agent repeats 'Step 1' forever on a blank page

  The browser opens but stays blank, and the log prints 'Step 1' over and over without doing anything.

  Last checked Oct 17, 2026#setup#loop2 steps →0
• LangGraphFixed (verified)

  Callbacks, tags or metadata bound with with_config are silently dropped

  Values bound on a graph with with_config (callbacks, tags, metadata, configurable) disappear or leak between runs when a per-call config is passed.

  Last checked Oct 17, 2026#config#state1 steps →0
• OpenClawFixed (verified)

  CLI runs out of memory on basic commands like gateway status after upgrading

  After upgrading to 2026.3.12, simple commands such as openclaw gateway status and openclaw doctor crash with JavaScript heap out of memory.

  Last checked Oct 17, 2026#regression#setup1 steps →0
• MCP Python SDKFixed (verified)

  ClosedResourceError in server logs when a request fails validation early

  An MCP server running Streamable HTTP logs ClosedResourceError when a request is rejected quickly, for example because of a wrong Accept header.

  Last checked Oct 17, 2026#race#transport2 steps →0
• Pydantic AIFixed (verified)

  Concurrent runs sharing an MCP server fail with a cancel scope RuntimeError

  Running several agent runs or evals at once against the same MCP server toolset eventually fails with RuntimeError: Attempted to exit cancel scope in a different task than it was entered in.

  Last checked Oct 17, 2026#race#transport2 steps →0
• LangChainFixed (verified)

  create_agent with a checkpointer exits early using last turn's structured response

  With a response_format and a checkpointer, a later turn ends early and returns the structured response saved from an earlier turn.

  Last checked Oct 17, 2026#state1 steps →0
• CrewAIFixed (verified)

  crewai reset-memories fails with 'No crew found'

  Running crewai reset-memories (for example with --knowledge or -a) inside a crew project fails with 'An unexpected error occurred: No crew found.'

  Last checked Oct 17, 2026#setup#error-messages1 steps →0
• CrewAIFixed (verified)

  crewai run crashes on Windows with UnicodeDecodeError 'charmap'

  On Windows, crewai run crashes right away with UnicodeDecodeError: 'charmap' codec can't decode byte.

  Last checked Oct 17, 2026#setup#regression2 steps →0
• Hermes AgentFixed (verified)

  Cron jobs longer than 60 seconds are marked 'Interrupted by shutdown' even though they succeeded

  Cron jobs whose delivery takes more than about a minute are recorded as failed with 'Interrupted by shutdown before terminal completion', though the work finished.

  Last checked Oct 17, 2026#race#error-messages1 steps →0
• Pydantic AIFixed (verified)

  Custom httpx client given to an MCP server is closed after one run

  After the first agent.run(), the httpx.AsyncClient passed to MCPServerStreamableHTTP is closed, and the next run fails with RuntimeError: Cannot reopen a client instance, once it has been closed.

  Last checked Oct 17, 2026#transport1 steps →0
• MCPFixed (verified)

  Elicitation fails on edge runtimes with 'Code generation from strings disallowed'

  Calling elicitInput from an MCP server on an edge runtime such as Cloudflare Workers throws EvalError during schema validation.

  Last checked Oct 17, 2026#setup#schema2 steps →0
• LangGraphFixed (verified)

  get_state() is wrong after a node calls interrupt() more than once

  After resuming a node that interrupts twice, get_state() doesn't match what the graph is really waiting on: next can be empty, or an already answered question is still listed as pending.

  Last checked Oct 17, 2026#state#messaging1 steps →0
• AutoGenFixed (verified)

  Group chat fails with 'model client not activated' when the manager uses a custom client

  A GroupChat with speaker_selection_method="auto" errors as soon as it tries to pick the next speaker when the manager uses a custom model client.

  Last checked Oct 17, 2026#config#messaging2 steps →0
• Hermes AgentFixed (verified)

  hermes update fails with 'RPC failed; HTTP 429'

  Running hermes update prints 'Failed to fetch updates from origin' with an HTTP 429 error.

  Last checked Oct 17, 2026#error-messages#transport2 steps →0
• Hermes AgentFixed (verified)

  hermes update fails with "cannot import name 'file_signature' from 'utils'"

  hermes update pulls new code and then dies with ImportError: cannot import name 'file_signature' from 'utils'.

  Last checked Oct 17, 2026#regression#setup2 steps →0
• CrewAIFixed (verified)

  Hierarchical crew answers end with stray triple backticks

  In a hierarchical crew, agent final answers often end with an extra ``` line.

  Last checked Oct 17, 2026#parsing#tool-calls2 steps →0
• CrewAIFixed (verified)

  import crewai takes 25 seconds or more

  Simply importing crewai takes 25 seconds or longer before any code runs.

  Last checked Oct 17, 2026#setup1 steps →0
• LangChainFixed (verified)

  Import fails with 'No module named tenacity.asyncio'

  Importing LangChain fails after the tenacity dependency is upgraded to 8.4.0.

  Last checked Oct 17, 2026#setup#regression2 steps →0
• n8nFixed (verified)

  Installing community nodes on Windows fails with 'spawn npm ENOENT'

  On Windows, installing a community node from Settings fails with 'spawn npm ENOENT'.

  Last checked Oct 17, 2026#setup#regression1 steps →0
• LangGraphFixed (verified)

  Interrupt inside a tool silently ends the run after upgrading

  After upgrading to the 0.5 line, calling interrupt() inside a tool no longer pauses for input; the run just exits with no error.

  Last checked Oct 17, 2026#regression#state2 steps →0
• AutoGenFixed (verified)

  json.dump of a team's save_state fails with 'Object of type datetime is not JSON serializable'

  Saving a team's state to a JSON file raises a TypeError, and the file is cut off at the first timestamp.

  Last checked Oct 17, 2026#state#error-messages1 steps →0
• CrewAIFixed (verified)

  JSONLoader reports a parse error for valid JSON files saved with a BOM

  A valid UTF-8 JSON file loaded through the RAG JSONLoader comes back unparsed, with parse_error set in its metadata.

  Last checked Oct 17, 2026#parsing#error-messages2 steps →0
• CrewAIFixed (verified)

  Large CSV knowledge source fails with an embedder token limit error

  Loading a large CSV through CSVKnowledgeSource fails with an embedder error about the token limit.

  Last checked Oct 17, 2026#config1 steps →0
• MCPFixed (verified)

  listTools() silently returns a short list when a server repeats a cursor

  listTools(), listPrompts(), listResources() or listResourceTemplates() return fewer items than the server has, with no error.

  Last checked Oct 17, 2026#transport#schema1 steps →0
• Hermes AgentFixed (verified)

  macOS Desktop update fails with 'Another Hermes update is already running'

  On macOS, every update started from the Desktop app's Update button fails with exit code 2, and the app keeps retrying every few minutes.

  Last checked Oct 17, 2026#setup#regression2 steps →0
• n8nFixed (verified)

  MCP Client Tool calls fail after upgrading because of an extra toolCallId argument

  After upgrading n8n, workflows that call tools through the MCP Client Tool node start failing.

  Last checked Oct 17, 2026#tool-calls#regression1 steps →0
• MCPFixed (verified)

  MCP servers launched with npx won't start on Windows without cmd /C

  On Windows, a stdio MCP server configured with command npx fails to start unless it's wrapped in cmd /C.

  Last checked Oct 17, 2026#setup2 steps →0
• AutoGenFixed (verified)

  MCP tool calls fail with 'unhandled errors in a TaskGroup'

  An AutoGen agent calls an MCP tool with the right query, but the call fails with 'unhandled errors in a TaskGroup (1 sub-exception)'.

  Last checked Oct 17, 2026#tool-calls#schema1 steps →0
• AutoGenFixed (verified)

  MCP tool errors show up as a list of empty {} objects

  When an MCP server tool raises an error, the agent gets back [{}, {}, {}, ...] instead of the error message.

  Last checked Oct 17, 2026#error-messages#tool-calls1 steps →0
• ContinueFixed (verified)

  MCP tools get a JSON object where their schema expects a string

  MCP tools such as write_file receive a parsed object for a string parameter (like file content that is JSON), so the call fails or writes the wrong content.

  Last checked Oct 17, 2026#tool-calls#schema1 steps →0
• MCPFixed (verified)

  MCP TypeScript SDK fails in CommonJS projects with ERR_REQUIRE_ESM

  Importing the MCP TypeScript SDK (directly or through an adapter) in a CommonJS project such as NestJS fails at runtime.

  Last checked Oct 17, 2026#setup1 steps →0
• Pydantic AIFixed (verified)

  new_messages() loses messages when a history processor trims the current run

  With a history processor that keeps only the last few messages, result.new_messages() returns just the last few messages of the run instead of everything the run added.

  Last checked Oct 17, 2026#state1 steps →0
• Pydantic AIFixed (verified)

  new_messages() returns the user prompt passed in message_history

  Calling agent.run() with message_history and no user_prompt makes result.new_messages() include the request from that history, even though it was input, not something the run produced.

  Last checked Oct 17, 2026#state#regression2 steps →0
• MCPFixed (verified)

  OAuth metadata ignores the path in the auth router's base URL

  With mcpAuthRouter mounted under a path such as /auth, the advertised OAuth endpoints and protected-resource metadata drop that path.

  Last checked Oct 17, 2026#config2 steps →0
• OpenClawFixed (verified)

  openclaw update fails with 'FICLONE: Operation not permitted' in an unprivileged container

  openclaw update stops in the validating phase at step updater-runtime-retention inside an unprivileged LXC container.

  Last checked Oct 17, 2026#setup#error-messages1 steps →0
• Pydantic AIFixed (verified)

  prepare_tools callback returning None silently hides every tool

  After a prepare_tools callback returns None, the agent runs with zero tools and answers in prose, with no exception, log or trace entry.

  Last checked Oct 17, 2026#tool-calls#error-messages2 steps →0
• browser-useFixed (verified)

  Program never exits when using Browser(keep_alive=True)

  With Browser(keep_alive=True), the script stops after the first task and asyncio.run() never returns.

  Last checked Oct 17, 2026#state#setup1 steps →0
• Hermes AgentFixed (verified)

  QQ Bot gateway stuck retrying: 'unexpected keyword argument is_reconnect'

  After an update, the QQ Bot gateway never comes back online and retries every 60 seconds.

  Last checked Oct 17, 2026#transport#regression2 steps →0
• Hermes AgentFixed (verified)

  Read-only MCP tools on untrusted servers are always held for approval

  On an MCP server set to trust: untrusted, even tools marked readOnlyHint: true wait for approval, and in non-interactive runs they are refused.

  Last checked Oct 17, 2026#tool-calls#config1 steps →0
• AutoGenFixed (verified)

  Redis chat cache never hits, so every request goes to the model

  With ChatCompletionCache backed by RedisStore, responses always come back with cached=False even when the key exists in Redis.

  Last checked Oct 17, 2026#state#config2 steps →0
• MCP Python SDKFixed (verified)

  ResourceWarning: Unclosed MemoryObjectReceiveStream from the MCP client

  Apps using the MCP Python client log ResourceWarning about unclosed MemoryObjectReceiveStream and MemoryObjectSendStream.

  Last checked Oct 17, 2026#state#transport1 steps →0
• LangGraphFixed (verified)

  Resuming one parallel interrupt in a subgraph re-runs the node that already finished

  A subgraph pauses on two parallel interrupts; answering them one at a time re-runs the node that was already answered, repeating its side effect.

  Last checked Oct 17, 2026#state#loop2 steps →0
• MCPFixed (verified)

  Reusing one McpServer in createMcpHandler leaks memory, then RangeError

  A server that returns the same McpServer for every request through createMcpHandler slowly grows in memory and eventually throws an uncatchable RangeError.

  Last checked Oct 17, 2026#state#regression2 steps →0
• browser-useFixed (verified)

  Run that hits max steps returns an empty history with no error

  When the agent runs out of steps it logs 'Failed to complete task in maximum steps', but the returned history list has no error entry.

  Last checked Oct 17, 2026#error-messages#loop2 steps →0
• Pydantic AIFixed (verified)

  run_sync() freezes when a sync output function calls another agent

  An agent run with run_sync sometimes hangs forever when its output function calls a second agent with run_sync.

  Last checked Oct 17, 2026#race#tool-calls2 steps →0
• browser-useFixed (verified)

  Second run after add_new_task fails with QueueShutDown

  The first task runs fine, but calling run() again after agent.add_new_task() raises bubus.service.QueueShutDown.

  Last checked Oct 17, 2026#state#messaging2 steps →0
• browser-useFixed (verified)

  send_keys with '+' types nothing instead of a plus sign

  Asking the agent to type a literal + (or press a shortcut like Control++) sends empty key events and nothing is typed.

  Last checked Oct 17, 2026#browser#parsing1 steps →0
• browser-useFixed (verified)

  Sensitive-data redaction garbles output when one secret is part of another

  redact_sensitive_string produces broken placeholders when a shorter secret value appears inside a placeholder written for a longer one.

  Last checked Oct 17, 2026#security#parsing1 steps →0
• MCPFixed (verified)

  Server keeps running request handlers after the client disconnects

  When a client disconnects mid-request, the TypeScript MCP server keeps running the in-flight handlers to completion.

  Last checked Oct 17, 2026#transport#state2 steps →0
• MCP Python SDKFixed (verified)

  Server replies with a numeric id when the client sent a string id

  A client sends a JSON-RPC request with a string id such as "1", and the Python server responds with the number 1, so the client can't match the reply.

  Last checked Oct 17, 2026#schema#transport1 steps →0
• MCP Python SDKFixed (verified)

  Server returns 400 instead of 404 for an expired session, so clients don't reconnect

  After a Streamable HTTP server restarts or drops a session, clients keep failing with 400 Bad Request instead of starting a new session.

  Last checked Oct 17, 2026#transport#config1 steps →0
• OpenClawFixed (verified)

  Slack stops answering after a few days with no error

  A Slack connection that worked for days stops responding to DMs and mentions, with no visible error.

  Last checked Oct 17, 2026#transport2 steps →0
• Pydantic AIFixed (verified)

  Sliding window compaction drops the current request when keeping one message

  With SlidingWindowCompaction(keep_messages=1, receipts=True), every time it trims, the model gets only the compaction receipt and never sees the prompt it is supposed to answer.

  Last checked Oct 17, 2026#state#config2 steps →0
• MCPFixed (verified)

  stdio MCP servers stay running after the client closes, piling up as zombies

  Each time the client window closes or restarts, the old TypeScript stdio MCP server process keeps running, and processes accumulate.

  Last checked Oct 17, 2026#transport#setup2 steps →0
• MCP Python SDKFixed (verified)

  stdio MCP tool that runs a subprocess hangs until timeout

  A tool on a stdio MCP server that starts an external script never returns; the call hangs until the client times out.

  Last checked Oct 17, 2026#transport#race2 steps →0
• MCPFixed (verified)

  Streamable HTTP client overwrites a custom Accept header

  A custom Accept header set in requestInit.headers is replaced by the SDK default, breaking endpoints that need a specific value.

  Last checked Oct 17, 2026#transport#config1 steps →0
• MCP Python SDKFixed (verified)

  Streamable HTTP server hangs when several requests stream at once

  SSE responses from a Streamable HTTP MCP server hang indefinitely when several requests are in flight or a response has several parts.

  Last checked Oct 17, 2026#race#transport1 steps →0
• Pydantic AIFixed (verified)

  Streamed run stops right after a tool call when the model sends empty text first

  With some local models, an agent run that uses tools stops right after the tool is called and returns nothing.

  Last checked Oct 17, 2026#parsing#tool-calls1 steps →0
• LangGraphFixed (verified)

  Streaming with streamMode 'messages' changes messages saved in checkpoints

  With streamMode: "messages", AI messages from a subgraph lose properties such as name when they're reloaded from the checkpointer.

  Last checked Oct 17, 2026#state1 steps →0
• LangChainFixed (verified)

  Summarization failure replaces the whole conversation with error text

  When SummarizationMiddleware's summary call fails, the conversation history is deleted and the error text is stored as the summary.

  Last checked Oct 17, 2026#state#error-messages1 steps →0
• Pydantic AIFixed (verified)

  system_prompt is skipped when the agent runs through a UI adapter

  An agent served through AGUIAdapter, VercelAIAdapter or to_web runs without its system_prompt, and functions decorated with @agent.system_prompt never run.

  Last checked Oct 17, 2026#config#state2 steps →0
• Hermes AgentFixed (verified)

  Telegram gateway won't connect: 'do_request' is read-only

  After pulling the latest Docker image, the Telegram gateway keeps failing to connect and retries in a loop.

  Last checked Oct 17, 2026#transport#regression2 steps →0
• Pydantic AIFixed (verified)

  Temporal retries an MCP tool forever when it raises ModelRetry

  In a Temporal agent, an MCP tool that raises ModelRetry, such as a fetch that hits a 404, fails the activity, and Temporal retries it endlessly instead of handing control back to the agent.

  Last checked Oct 17, 2026#loop#tool-calls1 steps →0
• MCP Python SDKFixed (verified)

  Tool call hangs until timeout when the server sends a bad response

  With the Streamable HTTP client, call_tool waits until timeout even though the server already finished, when the response can't be parsed.

  Last checked Oct 17, 2026#transport#error-messages2 steps →0
• LangChainFixed (verified)

  ToolRuntime isn't injected when a tool has a custom args_schema

  A @tool with args_schema set to a Pydantic model fails to receive its runtime argument (ToolRuntime or InjectedToolCallId).

  Last checked Oct 17, 2026#tool-calls#schema1 steps →0
• Pydantic AIFixed (verified)

  web_fetch_tool ends the run on a URL with an invalid hostname

  When the model asks web_fetch_tool for a URL whose hostname fails IDNA parsing, the run ends with httpx2.InvalidURL instead of letting the model try again.

  Last checked Oct 17, 2026#tool-calls#error-messages1 steps →0
• Hermes AgentFixed (verified)

  Windows installer hangs after the browser download reaches 100%

  On Windows, the installer stops responding during the browser install step after the download shows 100%, with no success or failure.

  Last checked Oct 17, 2026#setup2 steps →0
• AutoGenFixed (verified)

  'str' object has no attribute 'get' when the chat finishes

  The task output prints, then the run crashes with AttributeError: 'str' object has no attribute 'get'.

  Last checked Oct 16, 2026#schema#error-messages3 steps →0
• LangGraphWorkaround

  "tool_use ids were found without tool_result blocks" on a saved thread

  An agent with a database checkpointer fails on every new message in one thread with a tool_use/tool_result mismatch error.

  Last checked Oct 16, 2026#state#tool-calls3 steps →0
• AutoGenWorkaround

  Adding a second agent to the quick-start makes it loop forever

  After registering one more agent, the program never ends.

  Last checked Oct 16, 2026#messaging#loop3 steps →0
• LangGraphWorkaround

  Agent repeats the same tool call until it hits the recursion limit

  A prebuilt agent keeps calling tools step after step and only stops when GraphRecursionError is raised.

  Last checked Oct 16, 2026#loop#tool-calls5 steps →0
• n8nWorkaround

  AI Agent tools fail with 'schema must be type object, got None' after upgrading

  After upgrading, the agent fails before calling any tool.

  Last checked Oct 16, 2026#regression#tool-calls3 steps →0
• browser-useFixed (verified)

  Browser agent can't pick an option from a dropdown

  The agent clicks the dropdown, then fails or loops trying to choose an option.

  Last checked Oct 16, 2026#browser#regression4 steps →0
• browser-useUnverified

  Browser agent keeps emitting empty actions until max steps

  Sometimes the agent outputs an empty action every step until the step limit, even though it read the right data.

  Last checked Oct 16, 2026#schema#tool-calls3 steps →0
• browser-useWorkaround

  Browser agent keeps going after it already called done

  In headless mode with a custom structured-output controller, the run never ends after the task is complete.

  Last checked Oct 16, 2026#loop#schema3 steps →0
• LangGraphOpen

  Cancelled async tool leaves history invalid even with handle_tool_errors=True

  After a tool task is cancelled, the next model call fails with INVALID_CHAT_HISTORY.

  Last checked Oct 16, 2026#state#tool-calls3 steps →0
• browser-useOpen

  Click on a missing element is reported as success

  Actions continue as if a click or dropdown pick worked, but nothing happened on the page.

  Last checked Oct 16, 2026#error-messages3 steps →0
• Coding agent CLIUnverified

  Coding agent CLI Desktop "Run agent in WSL": every command fails with "Failed to create unified exec process"

  On Windows with "Run agent in WSL" on, every isolated command fails before it starts, even pwd. Commands you approve one by one still work. The .codex/tmp/arg0 folder often holds only Windows .bat helpers and no codex-linux-sandbox.

  Last checked Oct 16, 2026#setup#race5 steps →0
• Coding agent CLIUnverified

  Coding agent CLI on Linux fails with "mountinfo path is not absolute" after an update

  After a Desktop or VS Code extension update that bundled CLI 0.155.0-alpha.16, every isolated command and patch fails while the Bubblewrap command is being built. System bwrap still works.

  Last checked Oct 16, 2026#regression#setup4 steps →0
• Coding agent CLIOpen

  Coding agent CLI on Windows: setup fails with a sharing violation, every command fails

  After a Windows Desktop update, every local command fails before it starts. Browser tools may fail too. Rebooting does not help.

  Last checked Oct 16, 2026#regression#setup3 steps →0
• Coding agent CLIFixed (verified)

  Coding agent CLI VS Code extension drops submitted messages or leaves them queued

  After an extension update, pressing Enter clears the composer but the message never appears and the agent does not reply. Resending several times eventually works; an older queued prompt may re-run after a reload.

  Last checked Oct 16, 2026#regression#messaging4 steps →0
• Coding agent CLIOpen

  Coding agent CLI: "API Error: Connection closed mid-response"

  Mid-turn, the agent appears to be working, then the reply stops with a connection error and little or nothing usable. More common in long sessions with many tool calls, after compaction, or when one tool call writes a very large payload.

  Last checked Oct 16, 2026#transport#error-messages4 steps →0
• Coding agent CLIOpen

  Coding agent ignores "stop writing comments" rules in its project instructions file

  The agent keeps adding verbose comments even when the project instructions file and memory say not to. Reminders clean up one edit, then the habit returns.

  Last checked Oct 16, 2026#config3 steps →0
• AiderWorkaround

  Coding agent stuck on 'did not conform to the edit format'

  Every edit fails with 'SEARCH/REPLACE block failed to match' and the session can't continue.

  Last checked Oct 16, 2026#parsing#loop3 steps →0
• Hermes AgentWorkaround

  Error messages name the symptom, not the cause

  You fix what the error says, and it still fails. For example, a 'payment/credit error' when there is simply no API key.

  Last checked Oct 16, 2026#error-messages3 steps →0
• CrewAIWorkaround

  Every tool call arrives with empty arguments and fails validation

  Each tool call fails with 'Field required ... input_value={}' and the agent repeats until max iterations.

  Last checked Oct 16, 2026#tool-calls#schema3 steps →0
• MCPFixed (verified)

  Fetch tool can be steered to localhost or cloud metadata

  An agent with the fetch server can be told, or redirected, to read internal addresses.

  Last checked Oct 16, 2026#security3 steps →0
• MCPFixed (verified)

  Filesystem server says 'path outside allowed directories' on the first call

  The first tool call right after connecting is denied. A moment later the same call works.

  Last checked Oct 16, 2026#race#setup2 steps →0
• AutoGenFixed (verified)

  Graph flow errors with 'unable to find next speaker' when a loop has tools

  A graph with a reviewer→writer revise loop fails at runtime even though the edge exists.

  Last checked Oct 16, 2026#loop3 steps →0
• LangChainWorkaround

  JSON output parser returns None instead of raising on bad JSON

  A chain returns None for some malformed outputs, and later code fails far from the cause.

  Last checked Oct 16, 2026#parsing#error-messages3 steps →0
• MCPUnverified

  npx-based server shows zero tools when Node comes from nvm

  The server is configured but never connects. Servers built as .exe or binaries work.

  Last checked Oct 16, 2026#setup3 steps →0
• n8nOpen

  One bad tool call ends the whole agent run

  The run ends with 'Received tool input did not match expected schema' and the user gets nothing.

  Last checked Oct 16, 2026#tool-calls#schema3 steps →0
• LangChainWorkaround

  Parser fails when the model wraps JSON in ```JSON instead of ```json

  parse_json_markdown raises on output that is valid JSON inside a code fence.

  Last checked Oct 16, 2026#parsing2 steps →0
• LangGraphWorkaround

  Setting recursion_limit to 10000 silently does nothing

  with_config({"recursion_limit": 10000}) has no effect on a subgraph or agent.

  Last checked Oct 16, 2026#config3 steps →0
• MCP Python SDKWorkaround

  Stateless MCP server logs ERROR every time a client disconnects

  Two ERROR records appear whenever a client cancels a request, and alerts fire.

  Last checked Oct 16, 2026#error-messages#transport3 steps →0
• MCP Python SDKOpen

  Streamable HTTP client keeps reconnecting past its retry limit

  A long-running request never resolves. The client keeps reconnecting.

  Last checked Oct 16, 2026#transport2 steps →0
• LangGraphWorkaround

  Token limit exceeded because retriever output piles up in chat history

  A ReAct agent with a persistent checkpointer starts failing with token-limit errors after a few turns.

  Last checked Oct 16, 2026#state4 steps →0
• ContinueWorkaround

  Tool calls fail because arguments come back as XML-like text

  In agent mode, tool calls break with empty arguments or list/index errors when using a self-hosted compatible endpoint.

  Last checked Oct 16, 2026#tool-calls#serving3 steps →0