zero-lab-ai/skills

What it is

zero-lab-ai/skills is a repository of 56 SKILL.md agent skills from zero-lab.ai, grouped into 10 plugins. The scope is authorized defensive assessment, penetration testing, red teaming, detection engineering, purple teaming and engagement operations. The page does not list the individual skills or say which plugin holds which.

Preconditions / Affected

The README says the skills support authorized work only. They do not grant permission, expand scope or replace an organization's rules of engagement. The page describes skills that touch adversary behavior as built to measure exposure, preserve evidence and hand results back to defenders.

What to do

Install the whole cross-agent skills tree:

npx skills add zero-lab-ai/skills

For Claude or Codex workspaces, import from .claude-plugin/marketplace.json. It registers the ten bundles, and installing a plugin gives every skill in that bundle.

To take one capability, copy a single plugins/<bundle>/skills/<skill>/ directory into the location your harness scans. The README says nothing outside that directory is needed at runtime, so a single copy is self-contained. Each skill keeps its core instructions in SKILL.md. It also has agents/openai.yaml for ChatGPT display, and references, scripts, assets, tests and evaluations only where the capability needs them.

Read the skills before granting an agent access to sensitive systems. The README says agent skills are executable instructions and should get the same review as code.

Report vulnerabilities in a shipped skill or script privately to contact@zero-lab.ai, not through a public issue. SECURITY.md has the review and reporting policy.

Caveats

The make check release gate verifies that:

• plugin manifests agree with each other;
• frontmatter and UI metadata are complete;
• local paths resolve;
• shell and Python helpers pass linting;
• every discovered test suite passes.

It also checks that security skills reject common rationalizations that cause missed findings. That check covers repository hygiene. It does not show that a skill finds real issues.

The README states that script output does not authenticate evidence, authorize testing or prove a target is secure.