Tamper-evident controls for your agents.

C2PA-style tamper detection for agent control files like SOUL.md, TOOLS.md, SKILLS.md, and AGENTS.md — detecting unauthorized changes from internal, external, and agentic sources.

User phone

Hello World Agent

Load boundary

Before model context

The app tries to read an agent control file. Free2PA checks its signed receipt first.

1 App asksSay hello

↓

2 Gate checkssignature + file hash + trusted publisher

↓

3 Safe contextverification reaches the model

Free2PA verify console

Free2PA verify console ready

READY

Verification pending Waiting for the Hello World app to read its control file.

Signed file ready Free2PA verifies this file before its text can enter model context.

Signed SOUL.md

Loading SOUL.md…

Signed original loaded from receipt demo fixture.

What this page proves

The app reads an agent control file. Free2PA verifies the signed receipt before the agent can consume it. If the file still matches, Free2PA returns LOAD. If a critical control file was edited without a valid new signature, Free2PA restores the signed original first, then lets only that verified text enter model context. Signing the edit creates a new trusted receipt, so the edited SOUL.md can become the approved version.

Where to add this in your app

Put loadVerifiedFile(), the CLI, or MCP verify_asset exactly where your app reads prompts, skills, policies, or tool files. See this installed in Hello World.