The weekly · Week 36 · 31 Aug – 6 Sept 2026
This week: agent skills for writing, science and editorial imagery, plus a lesson on guarding an agent's dangerous permissions.
4 picks
A writing skill for Agent Skills-compatible agents, with plugins for Claude Code, Codex, Grok Build and Antigravity. It repairs narrative structure in fiction and applies venue-matched rules to professional prose.
Image: GitHub
A library of 177 ready-to-use skills that lets AI agents work on science tasks, plus 100+ scientific databases across biology, chemistry, medicine and drug discovery. Works with Claude Code and Codex.
Image: GitHub
A skill that generates editorial print-style images in a single ink, using warm paper tones, halftone photography, deliberate negative space and restrained typography. It helps designers working with AI agents.
Image: GitHub
A hands-on write-up on protecting an agent that can trigger a production rollback from prompt injection. Conclusion: dangerous permissions need a verification layer outside the prompt, not instructions inside it.
Picked from public sources (GitHub, Hacker News, DEV, Lobsters, Reddit, Product Hunt) by our daily scout. Each summary uses only what the source itself says; GitHub stars are as checked on the date shown.
