oCMS 0.31.0 keeps supported editor formatting, searches readable page text, and warns about removed markup. What changes for editors, and what operators should check before upgrading.
Tampered 2.1.312/2.1.313 builds of the Claude Code extension in Cursor's marketplace planted an SSH key and persistence. Who's affected, and what to check.
UTC timestamps, mandatory production MFA, and health-aware deployments: what oCMS 0.29 and 0.30 change for developers running real sites.
Five Claude Code presets, four real task types, 60 runs: the light presets passed every check at a tenth of the cost of Opus at high effort. How the harness, the graders and the fairness controls were built.
oCMS now ships a read-only MCP server: ten typed tools let Claude Code, Cursor and other agents search and read site content, with drafts hidden and every API key policy enforced.
OpenAI's GPT-5.6 (Sol, Terra, Luna) is a serious release — and that's good for everyone: better per-task quality, flexible token budgets, and a stronger Codex + Claude Code duo.
A hands-on take on Claude Fable 5: at the same effort level it's faster than Opus 4.8 with noticeably better code quality — but roughly a third of security audits get blocked by its widened cyber classifier and rerouted to Opus 4.8.
US export controls on Claude Fable 5 and Mythos 5 have been lifted; Fable 5 returns July 1 with a tighter cyber classifier (more false positives, Opus 4.8 fallback), a new industry jailbreak-severity framework, and deeper US government collaboration.
Anthropic's new Mythos-class model, Claude Fable 5, is its strongest coding and agentic model yet — cheaper than the preview, with a new safeguard layer that can fall back to Opus 4.8.
AI did not kill engineering. It killed the illusion that typing code was the hard part. In my Claude Code + Go + oCMS workflow, the human job moved upstream: intent, constraints, architecture, verification, and context discipline.