Claude Code session leak bug surfaces, llama.cpp and InvokeAI ship updates
A GitHub issue flagging potential session and cache leakage in Claude Code hit 307 points on HN. Meanwhile llama.cpp and InvokeAI both pushed new releases worth grabbing.
A GitHub issue flagging potential session and cache leakage in Claude Code hit 307 points on HN. Meanwhile llama.cpp and InvokeAI both pushed new releases worth grabbing.
A GitHub issue on the Claude Code repo — titled 'Potential session/cache leakage between workspace instances or consumer accounts' — racked up 307 points on Hacker News this week. If you use Claude Code across multiple workspaces or share an account, this is worth reading before you assume your context is isolated.
The issue (GitHub #74066) raises the possibility that session state or cached data bleeds between separate workspace instances. Anthropic has not yet posted a public fix or a confirmed root-cause analysis as of July 5. If you run Claude Code in a professional context — client work, anything with sensitive code — watch this thread. The safe move right now is to avoid assuming workspace isolation is airtight.
llama.cpp tagged build b9873 on July 4. No dramatic feature announcement, but llama.cpp ships fast and each build tends to include quantization fixes and backend performance work. If you run local models on Apple Silicon or CUDA, pulling the latest build is low-risk and usually worth it.
InvokeAI hit v6.13.5. InvokeAI is a local Stable Diffusion front-end aimed at non-developers — it has a node-based canvas and a one-click installer. A point release like this typically patches bugs from the previous minor version. If you're on v6.13.x, update.
The Model Context Protocol servers repo also tagged a release dated 2026-07-04. MCP servers are the connectors that let Claude and other agents talk to external tools — databases, APIs, file systems. If you've built or are building an MCP integration, check the changelog for any breaking changes.
Command and Conquer Generals was ported natively to macOS, iPhone, and iPad using a framework called Fable. The repo is on GitHub. This is not AI news, but it is a strong example of what a solo or small-team port looks like in 2026 — relevant if you're thinking about bringing an old desktop app to mobile.
1. If you use Claude Code professionally: read GitHub issue #74066 today and decide whether to pause multi-workspace use until Anthropic confirms a fix. Link your team to the thread.
2. Pull the latest llama.cpp (b9873) and run a quick benchmark on a model you already use locally — even a 7B GGUF. Note whether inference speed changed. It takes 20 minutes and gives you a real baseline.
3. If you want to connect Claude to a tool you use (Notion, a local database, a REST API), browse the MCP servers repo for an existing connector before building one from scratch. The July 4 release may have added one you need.
Confirmed: GitHub issue #74066 on the Claude Code repo exists and reached 307 HN points, flagging potential session/cache leakage between workspace instances.
Not independently verified by CBW: We have not reproduced the session leak ourselves. The severity, scope, and whether Anthropic has a fix in progress are unknown as of publication.
Confirmed: llama.cpp build b9873, InvokeAI v6.13.5, and modelcontextprotocol/servers 2026.7.4 are real tagged releases on GitHub.
Worth noting: The HuggingFace model uploads (Qwen3.6-40B, Huihui-Qwythos-9B, Krea-2 ControlNet) are community GGUF quants and fine-tunes — not official model launches. We skipped them as lead items for that reason.
Worth noting: The C&C Generals macOS/iOS port is a community project, not an official EA release. Legal status of distributing the game assets is unclear.
Source: Claude Code GitHub issue #74066 — session/cache leakage — https://github.com/anthropics/claude-code/issues/74066
Source: llama.cpp release b9873 — https://github.com/ggml-org/llama.cpp
Source: InvokeAI v6.13.5 release — https://github.com/invoke-ai/InvokeAI
Source: Model Context Protocol servers 2026.7.4 — https://github.com/modelcontextprotocol/servers
Source: C&C Generals macOS/iOS port (Fable) — https://github.com/ammaarreshi/Generals-Mac-iOS-iPad/tree/main
Get tomorrow's best AI project in your email. With a guide that works. Free. No spam.