Problemsby GlyphAnthropic upgrades break working setups
27 posts from 22 people21d activeSteady, +1 this week
Posts per day
The posts behind it
HTMLHeaderTextSplitter loses linked section headings, breaking section-filtered retrieval
### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...
[BUG] Anthropic tool_search given as a dict is silently disabled
> Label: `llm-generated` - this report was produced with AI assistance (ZCode agent) and human-reviewed. Please apply the `llm-generated` label. ## Problem When `tool_search` is passed as a dict (e.g. from a YAML agent config, where the `llm:` block arrives as plain dicts), the...
`TextSplitter._merge_splits` does not warn when the last chunk exceeds `chunk_size`
### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...
[Bug]: content filter MASK streaming sends held-back text after the tool call chunk, agent clients drop it
## Description With a `litellm_content_filter` guardrail running as `post_call` in MASK mode, a streamed chat completion where the model writes some text and then calls a tool reaches the client out of order. The client gets the start of the text, then the `tool_calls` delta,...
create_agent ends mixed batches early when a non-direct tool is registered at runtime
### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...
core: InMemoryVectorStore returns arbitrary documents for negative k
### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...
core: InMemoryVectorStore returns arbitrary documents for negative k
### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...
core: truncated tool-call arguments are parsed leniently after the stream ends, producing a valid-looking tool call
### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...
RouterRunnable batch ignores return_exceptions=True for unknown routes
### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...
`trim_messages(strategy="last", allow_partial=True)` can return messages that exceed `max_tokens`
### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...
Streaming completion reports finish_reason "stop" when the stream ends without any finish reason
### Description When a streamed chat completion ends cleanly before the provider sends a finish reason, `litellm.completion(stream=True)` still emits a last chunk with `finish_reason="stop"`. The provider never sent `stop`, so a caller can't tell a stream that was cut short from...
count_tokens_approximately omits tool calls when AIMessage content is a list
### Submission checklist - [x] This is a bug, not a usage question. - [x] The title describes the issue. - [x] I searched existing issues and PRs and did not find this omission. - [x] The self-contained reproduction below fails in LangChain. - [x] This reproduces on the latest...
count_tokens_approximately omits tool calls when AIMessage content is a list
### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...
count_tokens_approximately omits tool calls when AIMessage content is a list
### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...
XMLOutputParser.parse() removes valid elements with encoding attributes followed by newlines
### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...
`with_retry().batch()` returns another input's result when only some inputs recover on retry
...through `with_retry()`, and one input recovers on a later attempt while another input keeps failing until retries run out, the failing input is given the recovered input's output. In the example, "b" never succeeds, but `batch` returns "A" (the output of "a") in its slot. With...
[Bug]: Anthropic safeguards are dropped or sent without the required beta
## Scope and evidence This report tracks behavior in the pinned v1.100.0 baseline. Existing upstream PR references are listed below; this is not a claim that the problem remains unfixed in current releases. Please associate with the existing fix or close as duplicate if...
merge_message_runs raises for ChatMessage roles and FunctionMessage names that differ
### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...
core: Tee leaves its source open when a child is never started
### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...
`on_tool_error` does not receive the caller's `**kwargs` that `on_tool_start` and `on_tool_end` do
### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...
core: empty `group_ids` bypasses validation and filtering in `InMemoryRecordManager`
### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...
UsageMetadataCallbackHandler silently drops usage_metadata when model_name is missing
### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...
Token budget middleware that checks in with the user instead of hard-stopping
...as tool names or a predicate, so tools like `ask_user` or `save_progress` are never blocked by the budget and calling them does not end the run. - **Safety net.** If the model still calls a restricted tool after the check-in, the call is executed and the next `before_model` ends...
BaseLLM cache ignores per-call generation options and replays the wrong completion
### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...
500 "no user query found in messages" for tool-only turns on /v1/messages and /v1/chat/completions (no truncation required)
...through the model template like any other turn. Requiring a literal user text turn and failing with 500 otherwise breaks spec-compliant agent clients (Anthropic/OpenAI SDKs) whenever they send a follow-up turn that only returns tool output -- and permanently so once truncation...
ContextEditingMiddleware blocks the event loop with model-based token counting
### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...
`HumanInTheLoopMiddleware`: cross-tool edit bypasses the target tool's `interrupt_on` policy
### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...
Companies and products named
About this problem
Evidence
27 posts from 22 people in 4 places, about 3 a week over 64 days. Mostly on langchain-ai/langchain, BerriAI/litellm, ollama/ollama. Tools named alongside: OpenAI, Ollama, LangChain, Perplexity.
Frustration Frustration 0 of 3· Seen on GitHub issues
Summary, model-written
Anthropic users encounter bugs where tool execution budget controls fail and LangChain integration errors occur across multiple submissions.
How it was grouped
Posts that state a pain and match the "breaking-changes" rule about Anthropic. First post Sep 3, 2026, latest Nov 5, 2026. Corroborated: 3 or more posts from 2 or more people or places. Method
The same problem elsewhere
GitHub upgrades break working setups
87 posts from 75 people on GitHub issues, Hacker News, Stack Overflow.
Upgrades and new releases break working setups
76 posts from 62 people on GitHub issues, Reddit, App Store reviews.
Claude Code upgrades break working setups
67 posts from 60 people on GitHub issues, Hacker News, Reddit.
Codex upgrades break working setups
65 posts from 58 people on GitHub issues, Reddit.
DuckDB upgrades break working setups
37 posts from 28 people on GitHub issues.
Claude upgrades break working setups
35 posts from 34 people on GitHub issues, Reddit, Hacker News.
ClickHouse upgrades break working setups
31 posts from 20 people on GitHub issues.
PyTorch upgrades break working setups
26 posts from 21 people on GitHub issues, Reddit.
ChatGPT upgrades break working setups
26 posts from 26 people on GitHub issues, Community forums, App Store reviews.
OpenAI upgrades break working setups
25 posts from 25 people on GitHub issues, Reddit, Community forums.
Open WebUI upgrades break working setups
25 posts from 2 people on GitHub issues.
Gemini upgrades break working setups
24 posts from 24 people on GitHub issues, Community forums, Reddit.
Other Anthropic problems
Anthropic users keep filing bugs
Top post: "`HTMLHeaderTextSplitter` raises `ValueError` on init for the non-heading tags its splitter already supports", one of 36 posts from 33 people on GitHub issues, App Store reviews, Reddit.
Anthropic rate limits and usage caps block work
12 posts from 9 people on GitHub issues, Reddit, Community forums.
Trouble with Anthropic core
Top post: "core: parse_json_markdown returns the wrong value for a json5 fence and fails on other fence tags", one of 8 posts from 6 people on GitHub issues.
Anthropic token costs are hard to predict
6 posts from 6 people on Hacker News, GitHub issues, Reddit.
Trouble with Anthropic cache
Top post: "LLM cache stores a generation that fails structured-output parsing and then replays it", one of 6 posts from 6 people on GitHub issues.
Trouble with Anthropic messages
Top post: "/v1/messages: client disconnect before the first chunk leaves the upstream LLM request running", one of 6 posts from 6 people on GitHub issues.
Anthropic fails to return valid structured output
5 posts from 5 people on GitHub issues.
RAG on Anthropic returns irrelevant or missing context
5 posts from 5 people on GitHub issues, Reddit.
Trouble with Anthropic response
Top post: "mistralai: an error event inside a chat stream is silently dropped, and the truncated output is returned as a", one of 4 posts from 4 people on GitHub issues.
Anthropic runs out of context on real work
4 posts from 4 people on GitHub issues.
Anthropic pricing is too high or keeps changing
3 posts from 3 people on Hacker News, Reddit.
Prompt injection makes Anthropic features unsafe
3 posts from 3 people on GitHub issues, Hacker News.
History