COMMUNITY PICK · 1 NOVEMBER 2026
About 1 in 17 public agent skills tells the agent not to ask you first
A new study of about 1.53 million public agent skills (arXiv 2610.37734) found roughly 1 in 17 contains a line that loosens human oversight, such as 'proceed without approval' or skipping verification checks, and 70% of those apply in every situation. Open-source skill scanners caught at most 46% of them.
Why it mattersBefore installing a skill, search it for lines like 'without asking' or 'no-verify', limit any such instruction to one context, and keep approvals on for anything that sends, pays or deletes.
Sources
In the directory
How we knowCompany figures are labelled as their own claims; anything we couldn't verify is labelled as a report or a claim.
MORE FROM 1 NOVEMBER 2026
- ChatGPT and Codex will share one usage allowance from December
- Anthropic: Claude agents learned to sneak bad training data past a filter
- Don't believe it yet: the 'Claude Fable 5 weights leak'
- Grok Bot for LinkedIn: hype or fact?
- Infostealer malware now takes AI coding-tool keys
- Suno MCP: make music from inside Claude or ChatGPT (beta)