Probata.
Curated & reviewed

Claude agent seeds you can actually trust.

A “seed” is a single SKILL.md file that gives your Claude agent a new capability. Probata is a small, curated library of seeds — read, reviewed, and security-scanned before they’re published.

The free anchor is a prompt-injection scanner. Run it on this library too.

The seeds

Three free at launch. Every one is reviewed — and the first one lets you verify that yourself.

Flagship · security

prompt-injection-finder

Scans SKILL.md, CLAUDE.md, and other instruction files for prompt-injection attempts. Tiered findings with line-quoted evidence. Runs entirely locally — nothing leaves your machine. Built so the act of scanning can’t itself injure you.

Build

cofounder-agent-builder

Interviews you about a domain you want a persistent agent for, then generates a CLAUDE.md operating model, folder layout, starter artifacts, and a cadence. Turns “I want Claude to be my [X]” into a real working project.

Context

context-builder

Builds a single context.md reference card for your small business — product, customers, pricing, metrics, risks — so every future agent you spin up starts from the same baseline instead of re-learning your business each session.

Audit before you trust — including us

Trust shouldn’t be a vibe. A complete posture has three legs, and three short guides cover them.

Harden first prevent — shrink the blast radius before you handle anything untrusted. The worst targets aren’t files on disk — they’re your live, logged-in connectors.
Safely audit detect — quarantine → static scan → classify → verdict → least-privilege. Built so the act of scanning can’t injure you.
If compromised respond — contain → rotate → sweep → restore. The failsafe that works isn’t wiping the machine; it’s revoking what was exposed.

How to use a seed

Two ways in.

Install the whole library as a plugin (once it’s live in the community marketplace):

/plugin marketplace add anthropics/claude-plugins-community
/plugin install probata-seeds@claude-community

Skills then invoke under the namespace, e.g. /probata-seeds:prompt-injection-finder.

Or copy a single seed: download its SKILL.md, drop it in your project’s .claude/skills/<seed-name>/ directory, scan it first, then invoke it. New to local agents entirely? The full start-to-finish guide is the walkthrough.