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.
Next.js dev server (Turbopack) fails to match routes that are children of a dynamic segment nested inside route groups, returning instant 404 for the entire session. The issue is probabilistic, occurs under module graph load, and is repaired by any file event under app/. Production builds are unaffected.
InMemoryRecordManager treats an explicitly provided empty list for group_ids as if it were None because of truthiness checks. This causes update() to skip length validation and store records with group_id=None, and list_keys() to return all keys instead of an empty result when group_ids=[] is passed. The async methods have the same issue.
The on_tool_error callback does not receive the arbitrary keyword arguments (e.g., tenant, trace_id) that were passed to tool.invoke, while on_tool_start and on_tool_end do. This inconsistency causes loss of critical context during error handling.
In langchain's retry middleware, `should_retry_exception` checks `callable(retry_on)` before `isinstance`, treating exception classes as callables. Passing a single exception class (e.g. `TimeoutError`) invokes the class with the exception, returning a truthy exception instance, causing every exception to be retried regardless of the intended filter.
The helper function _get_data_content_block_types() reconstructs its result from get_args and get_type_hints on every call, even though the result is immutable at runtime. This is called repeatedly by is_data_content_block, leading to ~23ms CPU per ChatOpenAI payload for a 204-message history, blocking the event loop in async applications.
In Next.js 16.3.6 on Vercel, background cache regeneration started during streaming is not tracked by waitUntil, causing it to be aborted when the serverless function exits, resulting in failed revalidations.
langchain-core 1.6.6 contains two type contract inconsistencies: import_attr accepts package=None but raises TypeError due to always using relative imports, and AIMessage/AIMessageChunk.content_blocks return list[str] when content is a list of strings, violating the declared return type list[ContentBlock].
Next.js middleware can return a response with a relative Location header (as allowed by HTTP spec), but the internal adapter passes this value directly to the NextURL constructor, which requires an absolute URL, causing an 'Invalid URL' error.
The bare URL pattern in detect_url incorrectly matches patterns like 'Node.js/Express', '1.5/2.0', and 'python3.12/site-packages', leading to redaction or blocking of non-URL content.
In Next.js 16, lowering watchpack's aggregateTimeout from 200ms to 5ms causes the dev server's initial page scan to fire aggregation callbacks before the scan completes. The first partial aggregation builds the dynamic route table from an incomplete page list, so dynamic pages-router routes return 404 until a later aggregation rebuilds it.
Starting with 16.3.8, the Next.js image optimization server fetches remote images using ImageHttpAgent/ImageHttpsAgent without proxyEnv, causing direct connections that ignore HTTP_PROXY/HTTPS_PROXY and global undici dispatchers. In proxied environments, all remote image requests time out with 504.
Supabase Studio's Exposed functions view computes role access from explicit ACL entries and joins pg_roles, ignoring PostgreSQL's PUBLIC pseudo-role (grantee OID 0). Functions with default PUBLIC EXECUTE are reported as revoked for anon/auth/service_role even though PostgREST allows execution.
When a client component inside a Suspense boundary uses synchronous IO (e.g., Math.random or Date.now) during server rendering, the resumed segment omits the host elements rendered before the sync IO call. This causes a hydration mismatch (React error #418) and the boundary's fallback is replaced with an incomplete subtree.
When a server component calls notFound() inside a dynamic route, the SSR response contains an empty HTML shell with <html id="__next_error__"> instead of rendering the root layout and not-found.tsx content. This occurs because the not-found error is thrown during server rendering and caught by the outer app-render catch, which generates a minimal error document rather than re-rendering the tree with the not-found boundary active.
When a module using next/dynamic is imported by both Pages Router and App Router, ReactLoadablePlugin writes the App Router webpack module id into react-loadable-manifest.json, causing Pages Router client bundle to request the wrong dynamic chunk id and fail hydration with React error #418.
Client-side navigation to ISR pages can serve stale RSC payloads from browser cache without contacting the server, leading to mixed module versions and runtime errors after redeployment.
User reports that all saved SQL queries disappeared from Supabase Studio after a Supabase service incident and page refresh, but database functions remain. This suggests queries were stored in a non-persistent or less redundant location compared to functions.
When multiple SWC plugins are configured and Turbopack's persistent file system cache is enabled (default since Next.js 16.3.0), a restored cache can pair plugin modules with the wrong configuration objects, causing SWC plugin panics and build/dev failures.
File-system-cache `set()` omits the `.rsc` file for PPR routes, causing `get()` to treat the entry as missing after restart or eviction. This results in fallback shell re-rendering and stale RSC payloads for client navigation, leading to data inconsistency.
NomicEmbeddings.embed_image passes self.vision_model=None unconditionally to nomic.embed.image, overriding the SDK's default 'nomic-embed-vision-v1.5'; the SDK's response model assertion then fails with an empty AssertionError. The constructor overloads also omit vision_model, causing mypy call-overload errors.
In langchain-core v3 streaming, tool-call argument deltas are emitted as full snapshots of accumulated state, leading to O(n^2) event sizes for large arguments. This causes severe performance degradation and memory overhead in LangGraph Server and any consumer of the event stream.
Constructing ChatOllama, Ollama, or OllamaEmbeddings with a non-empty `client_kwargs` dict mutates that dict and its nested `headers` dict by injecting an `Authorization` header derived from base_url credentials. This leaks credentials into caller-owned objects and causes credentials from one URL to be sent to other hosts when the same dict is reused.
When a route is marked `dynamic = 'force-dynamic'`, fetches with `next: { revalidate: false }` are treated as uncached (revalidate: 0) due to a truthiness check in patch-fetch.ts, so each request hits the origin instead of using the Data Cache.
Creating or deleting a Send Email hook in the Supabase dashboard silently nullifies all custom SMTP fields and resets rate_limit_email_sent to default, leading to complete silent auth email failure when the hook is later disabled.
Supabase Studio's database password inputs (Reset database password and new project database password) only set autoComplete="off", which password managers like LastPass ignore. This allows the user's Supabase account password to be autofilled into the database password field, creating a credential reuse vulnerability and potential accidental database password exposure.
In Next.js 16.4.0 dev with Turbopack, Pages Router pages crash during hydration because required client modules are not included in any chunk. The regression was introduced by #97664, which seeded availability info from the _app chunk group into page chunk groups; in dev each entry has its own module graph, so availability items from _app are wrongly treated as already available, causing modules like a-page-only and b-boundary to be skipped.
On cold load with a delayed client component chunk, Next.js App Router repeatedly retries root hydration causing thousands of empty commits and significant main thread time. The issue is triggered by `useDeferredValue` calls in `Head` and `InnerLayoutRouter` that pass the current RSC value as the initial value even when no prefetch data exists during initial hydration. This leads to React scheduling deferred work that commits an empty tree, making hydration eligible to retry in a loop.
When a page calls notFound(), Next.js serves a minimal error shell (`__next_error__`) in the server HTML, omitting the root layout, `<html lang>` attribute, and the nearest not-found UI. This breaks accessibility, no-JS clients, and dynamic root layouts, and is inconsistent with the hydrated result. Affects App Router across multiple Next.js versions.
With default experimental.turbopackSharedRuntime enabled and cacheComponents true, the prerendered shell includes inline scripts for chunk bootstrap and performance timing that lack a nonce attribute, causing browser to block them under a strict nonce-based CSP. This prevents client-side hydration and interactivity.
The @next/codemod agents-md command computes the docs root path relative to the current working directory rather than the directory of the `--output` AGENTS.md file. In monorepos where the output file is at a different location (e.g. repo root) and Next.js is installed only in a workspace package, the generated root path is broken, causing agent tools to read nonexistent documentation paths.
LangChain, LlamaIndex, and Pydantic
Pinecone, Chroma, Weaviate, Qdrant, and neighboring stores
Next.js, Supabase, and the surrounding app stack