Security2026-06-02 · source-backed

SkillHarm shows agent skills are a privileged, under-secured attack surface.

Issue

2026-06-02

Confidence

source-backed

Sources

Redaction

redacted

Story

The framework auto-constructs adversarial skills targeting every lifecycle phase: creation, distribution, deployment, execution (arXiv). The core problem is that agents implicitly trust and execute skill instructions. We solved this for packages with signatures, lockfiles, and scanning. The skills ecosystem has none of it. If you run a skill registry, treat every skill as untrusted code. Sandbox it. Verify integrity.

Source trail

arXiv

Entities

Sandbox SkillHarm Verify

Provenance

Canonical issue: Ramsay Research Agent — June 2, 2026
AI generated: no
Story unit: 2026-06-02-skillharm-shows-agent-skills-are-a-privileged-under-secured-attack-surface
Labels: source-backed, canonical briefing excerpt

Fetching from the wire…

SkillHarm shows agent skills are a privileged, under-secured attack surface.

Story

Related stories

OpenAI Confirms GPT-5.6 in Codex Has Been Deleting User Files

Anaconda buys Kilo Code to lock in the agent layer

Guillermo Rauch Just Open-Sourced the Architecture Behind v0's 3 Million Users

Vercel's seven-agent "software factory" now writes a third of merged PRs on the AI SDK

Vercel says coding agents now trigger over half of all deployments

Agent Plugins 1.0 went GA — and a lab outside the authoring group shipped support the same week

Vercel's CTO names the model that makes this generalizable: Kimi K3, open weights, no offensive-security safeguards, top of DeepSec Bench.

Vercel put AI Gateway and Sandbox microVMs behind Nous Research's Hermes Agent.

Source trail

Entities

Provenance

SkillHarm shows agent skills are a privileged, under-secured attack surface.

Story

Related stories

OpenAI Confirms GPT-5.6 in Codex Has Been Deleting User Files

Anaconda buys Kilo Code to lock in the agent layer

Guillermo Rauch Just Open-Sourced the Architecture Behind v0's 3 Million Users

Vercel's seven-agent "software factory" now writes a third of merged PRs on the AI SDK

Vercel says coding agents now trigger over half of all deployments

Agent Plugins 1.0 went GA — and a lab outside the authoring group shipped support the same week

Vercel's CTO names the model that makes this generalizable: Kimi K3, open weights, no offensive-security safeguards, top of DeepSec Bench.

Vercel put AI Gateway and Sandbox microVMs behind Nous Research's Hermes Agent.

Source trail

Entities

Provenance