SQUID

capgate vs Damn Vulnerable MCP: sandbox test results

The author (capgate maintainer) evaluated capgate — a compile-time capability-to-sandbox compiler — against the Damn Vulnerable MCP (DVMCP) teaching corpus of ten intentionally-broken MCP servers. For each challenge the author wrote an honest, minimal manifest, compiled it with capgate, and checked whether the emitted boundary stopped the attack. Results: capgate fully prevents one class (Challenge 3: excessive permission scope), meaning the declared fs read was compiled to a directory mount that made the private files unreachable. It meaningfully contains several other classes (token exfiltration, RCE, command injection) by egress allowlisting, read-only mounts, network disablement, and IP-blocking, but it does not prevent model-layer attacks like prompt injection or tool poisoning. The post documents precise compiler approximations (notes/unenforceable fields), reproduction steps using capgate@0.0.3, and the practical limits of a capability-compiler as one layer in an LLM-security stack.

Custom Codex-Powered Code Review Bot for GitLab

An engineer built an internal Codex-powered code-review bot for GitLab that runs on a self-hosted OpenShift cluster and reuses an existing ChatGPT subscription. The service accepts GitLab webhooks, queues jobs in BullMQ (Redis), drives an OpenAI Codex CLI/SDK instance from a single worker, and posts summary and inline comments back to GitLab. The design emphasizes strict security and isolation: credentials are excluded from the Codex process tree, prompt-invoked shell commands run in a separate exec-sidecar without access to sensitive mounts, a seccomp BPF filter blocks AF_UNIX socket abuse, and a four-layer egress lockdown (pod-local CoreDNS, NetworkPolicy, and a Squid egress-proxy with FQDN allowlists) prevents data exfiltration. State is split across Redis (jobs), MariaDB (indexes of reviews/threads) and Codex session JSONL files on a PVC. The project is closed-source and was published 2026-05-10.

Google Tests Discover Feed on Homepage

DMEXCO 2023 showcased Petal Ads, Huawei’s mobile advertising platform, and its strategies for entering the Chinese market. The event featured a Petal Ads Masterclass led by Jaime Gonzalo, emphasizing the importance of reaching high‑quality Chinese audiences, and a panel with Alessandro Schintu (Huawei Petal Ads), Jan Heumüller (Ogury), and Anna Keller (Bayer). Petal Ads highlighted exclusive access to more than 500 million monthly active users in China via Huawei’s platform. YouGov’s 2022 ranking named Huawei the top brand in China, with a 47.4% foldable phone market share and over 115 million users with devices priced above €800. The DMEXCO presence underlined Petal Ads’ global reach, claiming ninefold growth since 2020 to 730 million MAU across 170 countries, and shown devices included the P60 Pro, Mate X3, and MatePad 11.5". The content was produced in collaboration with Huawei, and the booth and sessions featured industry discussions on international campaigns and brand building.