A sourced fix-playbook workspace for AI and agent engineers. Find the root cause, the affected version, the rollback-safe next step, and edge-case audits.
The Next.js contributing guide does not instruct contributors to run `pnpm build` before `pnpm dev`. On a fresh clone, `pnpm dev` fails due to missing generated artifacts, leading to cascading module not found and TypeScript errors.
Next.js generated ParamMap widens dynamic segment values to string, ignoring the return type of generateStaticParams even when dynamicParams is false. This causes components (layouts/pages) that declare narrower types such as 'en' | 'de' to fail type validation during build, despite correct runtime behavior.
`langchain-model-profiles` CLI `refresh` performs global string replacements (`true`->`True`, `false`->`False`, `null`->`None`) on a `json.dumps` result, corrupting quoted JSON keys and values that contain those substrings. The generated `_profiles.py` still imports successfully but with silently altered profile data.
`_ConfigurableModel.batch_as_completed` and `abatch_as_completed` incorrectly reject singleton tuple configs, causing an AttributeError despite the annotated `Sequence[RunnableConfig]` contract. The code checks only for `list`, not the abstract `Sequence`, so non-list sequences are passed to `ensure_config` unnormalized.
ImagePromptValue is marked serializable and dumps() succeeds, but loads() fails because the class is absent from both deserialization allowlists (legacy langchain and langchain_core namespaces) in langchain_core/load/mapping.py. Its three sibling prompt values all round-trip correctly.
RunnableParallel.get_input_schema merges step input schemas but fails to exclude the Pydantic v2 root model synthetic field 'root', causing the returned schema to reject every valid input that invoke accepts when any step is RunnablePassthrough or unannotated RunnableLambda.
`create_image_block` in langchain-core accepts base64 data without a `mime_type`, unlike its siblings `create_video_block`, `create_audio_block`, and `create_file_block`. This produces blocks that cannot be converted to valid provider payloads and violates the function's documented contract.
The Supabase Google login documentation contains a code snippet where hashedNonce is declared inside a .then() callback, making it inaccessible outside the callback. Users following the guide encounter a ReferenceError when trying to use hashedNonce, and may skip the nonce or pass the raw nonce to both Google and Supabase, reducing security.
In `_MultiModalMessagePromptTemplate.aformat`, the `DictPromptTemplate` branch incorrectly calls the synchronous `format()` instead of awaiting `aformat()`. This means any subclass overriding `aformat` is silently bypassed on the async path, breaking consistency with the sibling `StringPromptTemplate` and `ImagePromptTemplate` branches which correctly await their async variants.
Next.js 16.3.0 introduced a server actions runtime change that adds event listeners per action invocation without cleanup, causing Node's MaxListenersExceededWarning and potential memory leak. This is a regression from versions before 16.3.x.
The registry schema accepts arbitrary strings for `backendProviders`, allowing unsupported provider names to bypass validation. This can cause runtime errors or silent misconfiguration. The proposed change restricts values to the known supported providers using a Zod enum.
Static type checking fails for create_agent when a heterogeneous list of middleware objects with different state_schema types is combined with an explicit state_schema argument. Runtime behavior works correctly, but type inference collapses all middleware to AgentMiddleware[AgentState, ...], causing `no-matching-overload` errors.
Resolves three UI and state issues: incorrect last-child CSS selector causing visible extra grid line in learn app, stale project name input due to React defaultValue without key, and potential memory leak from uncleared timeout in lite-studio project page.
langchain-openai's `_create_usage_metadata` only extracts known keys (`audio`, `cached_tokens`, `reasoning_tokens`) from `prompt_tokens_details` and `completion_tokens_details`, causing provider-specific fields like `text_tokens`, `accepted_prediction_tokens`, and `rejected_prediction_tokens` to be silently lost.
The build process awaits `telemetry.flush()`, causing unnecessary delay before returning. Telemetry submission should be detached like in dev mode.
In Next.js with webpack builder and optimization.concatenateModules enabled, a client entry's enclosing ConcatenatedModule can be assigned module id 0. The manifest plugin checks truthiness instead of null, causing the client reference to be omitted and the build to fail during prerendering.
Following the Authentication with Cache Components guide with 'use cache: private' leads to one DB query per prefetched page link, multiplying user lookups. The documented pattern does not share cached user data across independent PPR requests.
When filtering objects in the Storage explorer using the search box, deleting one or more of the filtered objects causes the file list to reset and display the full unfiltered bucket contents, while the search input still shows the filter text. This leaves the UI in an inconsistent state and may mislead users into thinking matching items still exist.
pgMeta.tables.update (primary_keys path) and pgMeta.columns.update (is_unique:false and check paths) build DO $$ ... $$ blocks that embed identifiers via ident(), causing syntax error when any identifier contains the literal byte sequence $$.
In langchain-core, `merge_configs` calls `ensure_config` on each input, which injects `DEFAULT_RECURSION_LIMIT = 25` when absent. This makes an explicitly provided `recursion_limit=25` indistinguishable from the injected default, causing it to be silently removed from the merged config.
When filtering a numeric column by value 0 in Supabase Table Editor, the visible rows are filtered correctly but the row count still reflects the unfiltered table because the count query omits the filter due to a truthiness check on the converted value 0.
On Windows, FilesystemFileSearchMiddleware returns virtual paths containing native backslashes (e.g., '/src\\nested\\example.py') instead of the contract's forward-slash format, causing inconsistent and non-portable tool outputs.
On Windows, path validation in FilesystemClaudeTextEditorMiddleware and FilesystemClaudeMemoryMiddleware rejects valid nested virtual paths due to backslash vs forward slash mismatch in allowed_prefixes comparison.
Next.js reports an error that mentions 'searchParams' when the actual inaccessible dynamic data is 'params' inside a 'use cache' context. This misleads developers by pointing to the wrong property, causing confusion and wasted debugging time.
Three security regression tests in langchain-core's prompt loading tests unconditionally create symlinks, causing OSError [WinError 1314] on default Windows environments. This makes the core test suite fail for contributors without elevated privileges or Developer Mode, and the suite has no Windows CI to catch it.
The @next/codemod next-lint-to-eslint-cli transform hardcodes npx to run @eslint/migrate-config, which is incorrect for pnpm projects and can lead to package manager mismatch, lockfile pollution, or failures in strict environments.
When a blog category filter is active, paginated requests to /api-v2/blog-posts omit the category parameter, causing unfiltered posts to be appended to the list.
The max_length parameter is not applied when using the lightweight 'sentencizer' pipeline, causing spaCy's default limit of 1,000,000 characters to be used instead of the user-specified limit.
ChatDeepSeek only overrides sync `_generate` and `_stream` to translate `JSONDecodeError` from the OpenAI SDK into a DeepSeek-specific message. The inherited async `_agenerate` and `_astream` are not wrapped, so async callers still receive the opaque `json.decoder.JSONDecodeError: Expecting value: line 1 column 1 (char 0)`.
The parse_json_markdown function in langchain_core.utils.json uses a case-sensitive regex to identify fenced code blocks, causing parsing failures when models output fences with uppercase 'JSON' or 'Json'. This affects users of JsonOutputParser and other utilities relying on this function.
LangChain, LlamaIndex, and Pydantic
Pinecone, Chroma, Weaviate, Qdrant, and neighboring stores
Next.js, Supabase, and the surrounding app stack