← Back to Blog

OWASP Agentic Skills Top 10: The Layer Nobody Was Securing

2026-05-23·4 min read
SecurityOWASPAI AgentsClaude CodeSupply Chain

Skills are the part of the agentic stack that quietly got the most power and the least scrutiny. A skill is a folder of markdown and scripts that tells an agent how to run a multi-step workflow, and the agent follows it with your credentials, your filesystem, and your network access.

OWASP's Agentic Skills Top 10 (AST10) is the first structured threat model for that layer. The mental model the project uses is worth stealing: MCP defines how models talk to tools; AST10 defines what tools actually execute.

It's version 1.0, 2026 Edition, still in active development, led by Ken Huang with eight co-leads. The full v1.0 launch and OWASP flagship submission is targeted for Q4 2026.

The list

IDRiskWhat it means
AST01Malicious SkillsSkills built to do harm, shipped as helpful ones
AST02Supply Chain CompromiseA trusted skill's upstream gets owned
AST03Over-Privileged SkillsSkills asking for far more access than the job needs
AST04Insecure MetadataFrontmatter and manifests as a payload delivery channel
AST05Untrusted External InstructionsSkills fetching behavior from sources nobody verified
AST06Weak IsolationNo real sandbox between the skill and the host
AST07Update DriftThe skill you audited is not the skill running today
AST08Poor ScanningPattern-matching scanners that miss what matters
AST09No GovernanceNo inventory, no approvals, no audit log, no owner
AST10Cross-Platform ReuseOne malicious skill, ported across every registry

The ones that should worry you most

AST03: Over-Privileged Skills

The project documents 280+ leaked credentials traced to skills holding permissions they never needed. This is the classic problem, just with a new blast radius: a skill that only needs to read one directory gets handed the whole home folder, and every secret in it becomes reachable the moment anything goes wrong.

Ask of every skill you install: what is the smallest set of tools this needs? If the answer is "all of them," that's the finding.

AST05: Untrusted External Instructions

A skill that fetches a prompt, a config, or a script at runtime is a skill whose behavior you cannot audit. You reviewed the file on disk. The file on disk says curl something and follow it. Prompt injection stops being a theoretical concern and becomes the actual control flow.

AST06: Weak Isolation

The project cites 135K+ exposed OpenClaw instances as the illustration. The pattern generalizes: agent runtimes get deployed with default network exposure and no sandbox, then handed skills that assume a trusted environment.

AST07: Update Drift

You audit v1.2. The registry serves v1.9 tomorrow. Nothing in the default install flow tells you. The mitigation OWASP points at is boring and effective, e.g. immutable pinning and hash verification and almost nobody does it because skill registries make installation one command and pinning zero commands.

AST08: Poor Scanning

The uncomfortable one. Skills are mostly natural language, and instructions hidden in prose don't look like anything a regex was built to catch. A scanner that greps for eval and subprocess will pass a skill whose entire attack is a paragraph telling the model what to do next.

What to actually do about it

If you're installing skills from anywhere but your own repo:

  1. Read the whole skill before installing it. Every file, not just the SKILL.md. Including the ones that look like documentation.
  2. Pin versions. Commit the skill into your project rather than pulling it live, or verify a hash.
  3. Check what it reaches for. Network calls, filesystem writes outside the project, credential reads, shelling out.
  4. Keep an inventory. AST09 exists because most teams cannot answer "what skills are installed across our agents right now."
  5. Treat skill markdown as untrusted input. It's instructions to a model that acts on your behalf, same trust boundary as a web page you scraped.

Contributing

The project takes submissions for new AST entries through a web form on the project site, and the usual fork-branch-PR flow on GitHub for edits. They're also asking for anonymized research findings, mitigation improvements, translations, and tooling.

License is CC BY-SA 4.0, so you can build on it.

Project page: owasp.org/www-project-agentic-skills-top-10