AI & Agent Dev Bug SandboxDefensive Tech Intelligence

Drop in an error log

Name: AI & Agent Dev Bug Sandbox
Creator: AI & Agent Dev Bug Sandbox

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.

Active Intelligence Feed

high85%

ContextEditingMiddleware blocks the event loop with model-based token counting

When ContextEditingMiddleware uses token_count_method="model" through the async method awrap_model_call, it directly invokes the synchronous model.get_num_tokens_from_messages on the event-loop thread. For models like ChatAnthropic that perform a synchronous HTTP request for token counting, this blocks the entire event loop for the duration of each count, causing severe latency spikes and starving other async tasks.

V1.4.0 Langchain

high90%

Next.js 15.5.23 bundles vulnerable postcss and sharp dependencies

Running `npm audit` on a project created with Next.js 15.5.23 reports high severity vulnerabilities from transitive dependencies `postcss` and `sharp`. The Next.js package pins outdated versions of these packages, requiring a dependency bump to secure patch releases.

V15.5.23 Next.js

high75%

Regression: next/dynamic conditionally importing client components adds both to client bundle in Next.js 15

Next.js 15 preloads all dynamic imports inside client components, causing both conditional chunks to be sent to the client, increasing initial JavaScript size and TBT compared to Next.js 14. The issue requires a user-land workaround until an upstream fix is released.

V15.3.3 Next.js

high85%

Invisible navigation-menu viewport blocks clicks and text selection after dropdown closes

On supabase.com, opening and closing a navbar mega-menu leaves a permanently mounted container (forceMount) that overlays hero content and intercepts pointer events, preventing text selection and CTA clicks.

Next.js

high68%

Turbopack: "type": "commonjs" in package.json breaks every App Router source file

Setting `"type": "commonjs"` in a Next.js project using Turbopack causes build and dev failures because Turbopack enforces the declared module format on App Router source files (which are written as ESM) and on generated `.next-internal` server-action entries, even though Webpack ignores this field. The issue affects all App Router files and internal virtual modules, not just the `'use client'` proxy previously fixed.

V15.5.26 V16.3.5 V16.3.6 V16.4.0-Canary.43 Next.js

high92%

Provider errors returned in a 200 response body are never classified

langchain-openai fails to classify errors embedded in HTTP 200 responses from OpenAI-compatible gateways. Non-streaming invoke raises a bare ValueError, streaming raises an unclassified openai.APIError, and Responses API raises ValueError. This prevents callers from accessing is_retryable and other ModelError attributes, breaking retry logic and potentially causing silent degradation.

Active Intelligence Feed🟢🟢🟢

ContextEditingMiddleware blocks the event loop with model-based token counting

Next.js 15.5.23 bundles vulnerable postcss and sharp dependencies

Regression: next/dynamic conditionally importing client components adds both to client bundle in Next.js 15

Invisible navigation-menu viewport blocks clicks and text selection after dropdown closes

Turbopack: "type": "commonjs" in package.json breaks every App Router source file

Provider errors returned in a 200 response body are never classified

Bedrock Converse `content_blocks` drops redacted reasoning blocks in `output_version="v1"`

structuredContent is dropped from isError=True tool results in langchain-mcp-adapters

Parallel tool calls without index collide on positional fallback key and one is silently dropped

EvaluationResult.feedback_config silently drops unknown or partial dict fields due to coercion to FeedbackConfig

InMemoryVectorStore silently drops documents when embedding returns mismatched vector count

VectorStore.add_texts silently truncates texts when len(ids) != len(texts)

AddableDict silently drops data on type-incompatible merge (langchain_core.runnables.utils)

PromptTemplate._validate_input rejects non-dict Mapping inputs (ChainMap, UserDict, MappingProxyType) and silently misrenders single-variable templates

Agents silently fail to produce structured_response when model omits tool call

langchain-openai with_structured_output() silently drops previously bound tools, causing hallucinated responses

Using compaction causes Anthropic API failure when agent is invoked with print_mode="messages" or stream_mode="messages"

Optimistic routes wrongly predict rewritten pages, causing locale mismatch

Server Action responses bypass deploymentId version skew detection, crashing stale tabs with "module factory is not available"

PIIMiddleware only inspects newest message, allowing earlier PII to bypass detection

CSS url() imports resolve incorrectly when using matchResource (`!=!`) with mini-css-extract-plugin

partialPrefetching causes four renders per on-demand page, timer-retained render state leads to large memory spikes

HuggingFaceEndpoint drops token for dedicated Inference Endpoints (*.endpoints.huggingface.cloud)

FilesystemClaudeTextEditorMiddleware uses locale encoding causing Unicode errors and data corruption on non-UTF-8 systems

langchain-anthropic manual thinking validation misses Claude Opus 4.7/4.8 and Sonnet 5

Middleware rewrites are network-proxied when using `next start -H 127.0.0.1`, causing EPROTO 500 behind TLS-terminating proxies

HumanInTheLoopMiddleware async agent ainvoke fails with RuntimeError: Called get_config outside of a runnable context

next/image: cancelled image request hangs that image until server restart

Socket rewrites connection reuse can lead to a race condition and intermittent 500s

[langchain-groq] bind_tools ignores a known tool_calling=False model profile and sends a request that Groq rejects

ContextEditingMiddleware blocks the event loop with model-based token counting

Next.js 15.5.23 bundles vulnerable postcss and sharp dependencies

Regression: next/dynamic conditionally importing client components adds both to client bundle in Next.js 15

Invisible navigation-menu viewport blocks clicks and text selection after dropdown closes

Turbopack: "type": "commonjs" in package.json breaks every App Router source file

Provider errors returned in a 200 response body are never classified

Bedrock Converse `content_blocks` drops redacted reasoning blocks in `output_version="v1"`

structuredContent is dropped from isError=True tool results in langchain-mcp-adapters

Parallel tool calls without index collide on positional fallback key and one is silently dropped

EvaluationResult.feedback_config silently drops unknown or partial dict fields due to coercion to FeedbackConfig

InMemoryVectorStore silently drops documents when embedding returns mismatched vector count

VectorStore.add_texts silently truncates texts when len(ids) != len(texts)

AddableDict silently drops data on type-incompatible merge (langchain_core.runnables.utils)

PromptTemplate._validate_input rejects non-dict Mapping inputs (ChainMap, UserDict, MappingProxyType) and silently misrenders single-variable templates

Agents silently fail to produce structured_response when model omits tool call

langchain-openai with_structured_output() silently drops previously bound tools, causing hallucinated responses

Using compaction causes Anthropic API failure when agent is invoked with print_mode="messages" or stream_mode="messages"

Optimistic routes wrongly predict rewritten pages, causing locale mismatch

Server Action responses bypass deploymentId version skew detection, crashing stale tabs with "module factory is not available"

PIIMiddleware only inspects newest message, allowing earlier PII to bypass detection

CSS url() imports resolve incorrectly when using matchResource (`!=!`) with mini-css-extract-plugin

partialPrefetching causes four renders per on-demand page, timer-retained render state leads to large memory spikes

HuggingFaceEndpoint drops token for dedicated Inference Endpoints (*.endpoints.huggingface.cloud)

FilesystemClaudeTextEditorMiddleware uses locale encoding causing Unicode errors and data corruption on non-UTF-8 systems

langchain-anthropic manual thinking validation misses Claude Opus 4.7/4.8 and Sonnet 5

Middleware rewrites are network-proxied when using `next start -H 127.0.0.1`, causing EPROTO 500 behind TLS-terminating proxies

HumanInTheLoopMiddleware async agent ainvoke fails with RuntimeError: Called get_config outside of a runnable context

next/image: cancelled image request hangs that image until server restart

Socket rewrites connection reuse can lead to a race condition and intermittent 500s

[langchain-groq] bind_tools ignores a known tool_calling=False model profile and sends a request that Groq rejects

Ecosystem Topology

Agent Frameworks

Vector Databases

Toolchains

Active Intelligence Feed