TAG
#reliability
Stories connected through a shared idea, actor, technology or theme.
n8n 2.42.3 expands API-key scope administrationn8n 2.42.3 lets global members grant variable scopes to API keys and prevents task-runner termination after unhandled promise rejections. The editor also exposes Gateway credit promotions on community nodes.→Mastra adds default agent error recovery and durable execution fixesThe release enables three error processors by default, repairing incompatible history or prefills before retrying transient failures. It also adds a tool-call-resumed stream event, improves durable workflow fencing, and prevents duplicated evented steps during redelivery.→Cline SDK adds recoverable oversized MCP tool resultsInstead of losing tool output when it exceeds the model context, the SDK stores the complete result in a per-session cache and exposes a cline://cache URI. Agents can retrieve the remaining content by line range, with session-level size and expiry limits.→Google Antigravity CLI 1.2.5 improves subagent guidance and recoveryAntigravity CLI 1.2.5 automatically gives Markdown-defined agents that declare subagent invocation access to the available subagent roster and delegation instructions. It also preserves background task names, clears definitively rejected credentials, records externally interrupted commands as cancelled and fixes frozen thinking frames.→OpenCode 1.18.31 restores state across resumed agent sessionsVersion 1.18.31 restores model, effort, mode and reasoning-chunk boundaries when sessions are loaded, resumed or forked. It also makes remote configuration authentication failures visible during startup and exits with a failure status.→Cline SDK 0.0.83 introduces hub-managed Agent PluginsHub-managed plugins are discovered from explicit host paths, while workspace .agents/plugins directories are not scanned automatically and therefore cannot implicitly start repository-controlled MCP servers. The release also adds bounded retries for transient model failures, improves event-stream performance and prevents apply_patch Add File operations from silently overwriting existing files.→Cline Desktop 0.0.27 hardens credentials, provider routing and session recoveryThe release changes expired-login handling, prevents stored credentials from silently restoring after sign-out and adds actionable recovery controls. It also preserves provider-specific protocols for OpenCode Go models and retries with a stable session identity.→NestJS 12.0.2 repairs streaming, microservice and Fastify edge casesThe release keeps Socket.IO and WebSocket message streams alive when handlers throw and improves failure handling for rejected microservice reply subscriptions. It also fixes several core, path-normalisation and Fastify response edge cases and adds MQTT connection-attempt limits.→Letta Code 0.32.8 hardens approval recovery and agent messagingThe release retries message requests when a gateway returns an HTML error page and prevents malformed questions from crashing the approval interface. It also improves teleport ownership during approval recovery and consumes queued messages without sender holds.→Agent Deck 1.16.10 hardens multi-agent session delivery and restart handlingThe terminal session manager now delivers completions when transcript signals are stale and merges post-start saves with concurrent session detection. It also rechecks releases while open and verifies that a tmux session exists after spawning.→Claude Code fixes long-session permission regressions in read-only Git commandsThe patch fixes permission prompts that could reappear after a session had been running for a while. Anthropic identifies the issue as a regression introduced in version 2.1.269.→n8n 2.37.11 fixes workflow cleanup and external-secret failovern8n 2.37.11 ensures running jobs are cleaned up when a workflow run rejects. It also keeps a serving external-secrets provider active when its replacement fails.→Claude Code 2.1.263 fixes background-agent reliability failuresThe release fixes interrupted commands being shown as completed, prevents finished subagents from silently restarting during navigation, and addresses races when opening newly started sessions. It also improves background-session retention and worktree cleanup behaviour.→Cloudflare workerd hardens Python workers and hibernating WebSocketsThe release adds automatic Python-worker flags, synchronous try-read/write helpers and fixes for Pyodide dynamic library loading. It also improves hibernating WebSocket write serialisation and actor-fetch retries, while expanding test coverage for streams and queuing strategies.→Claude Code 2.1.252 fixes remote-session stalls and oversized failure outputVersion 2.1.252 fixes Remote Control sessions stalling after tools finish when the Claude service connection is degraded. It also prevents very large background-task failure messages from pushing conversations beyond the API request-size limit.→Codex CLI 0.152 adds MCP output caps and longer shell-command deadlinesThe release adds per-tool MCP output limits, package-style MCP server names and configurable shell-command timeouts beyond one hour. It also preserves approval context through compaction and rejects untrusted cloud-task URLs and redirects.→QA Cafe automates long-duration stability testing for broadband devicesThe new testing workflow repeatedly exercises broadband and Wi-Fi devices while monitoring CPU and memory through TR-181 over TR-069 or USP. It correlates resource trends with pass/fail results, logs and packet captures to expose leaks, exhaustion, crashes and gradual performance degradation.→MiniMax records a brief elevated-error incident for its LLM serviceMiniMax opened an official incident for elevated errors on its large-language-model service at 01:03 UTC on 30 August and marked it resolved at 01:06 UTC. The status page provides no cause or impact estimate, and similar recurring entries make broader significance uncertain.→GitHub's Copilot Cloud Agent status visibility degraded for hoursGitHub's status page said Copilot Cloud Agent tasks were still completing, but newly started tasks did not show ongoing progress and session output was delayed by about an hour. GitHub reported gradual recovery during the window and marked the incident resolved at 00:37 UTC on 21 August.→OpenCode v1.18.20 makes subagent failures resumable and provider retries explicitOpenCode 1.18.20 now surfaces failed subagent calls with resumable task IDs instead of returning empty results, and retries several network, capacity and temporary-provider failures. It also preserves provider-specific token settings and handles permission requests raised by subagents during opencode run.→Vals AI raises $40 million to expand model evaluationVals AI raised $40 million in Series A funding led by a16z, with participation from HRT Ventures, Next Ladder Ventures and existing investors. The company evaluates models on domain-specific tasks in areas including finance, law, healthcare and coding.→Claude API and Claude Code outage interrupted multiple Anthropic servicesAnthropic’s status API recorded the incident starting at 21:58:56 UTC on 16 August and resolved at 22:34:39 UTC. The affected components included the Claude API, Claude Code and the platform console, so both end-user and developer-facing surfaces were involved.→Claude Opus 5 and Sonnet 5 suffered a separate developer-facing degradationAnthropic opened a separate incident at 13:56:41 UTC for elevated errors on Claude Opus 5, later reporting that Sonnet 5 was degraded as well. Claude.ai, the API, Claude Code and Claude Cowork were marked operational again at 15:29:25 UTC.→Cloudflare reported elevated Durable Objects errors in SingaporeCloudflare recorded the incident at 07:47:04 UTC and said Durable Objects and downstream services in Singapore were returning increased errors. The provider resolved the incident at 08:58:06 UTC.→SQLite WAL-reset bug explains Tailscale outagesThe bug can occur when multiple connections write to and checkpoint the same SQLite database in WAL mode at nearly the same time. SQLite says the issue was present from version 3.7.0 through 3.51.2 and is fixed in newer releases.→RobustifAI targets deployment failures beyond benchmark scoresThe workshop takes place at IJCAI-ECAI on 15 August and covers robustness across models, deployment and user interaction. Its agenda includes distribution shift, adversarial inputs, tool use, human interaction, monitoring and assurance.→AIM4DK frames disappearing knowledge as an AI reliability problemThe workshop is being held at IJCAI-ECAI on 15 August and focuses on AI modelling for disappearing knowledge. Its stated concerns include data versioning, temporal degradation, reproducibility and the loss of sources that support model outputs.→