Content contribution: Lesson 18 — Securing AI Agents with cryptographic receipts

[tomjwxf]

I noticed Lesson 18 (Securing AI Agents) is listed as "Coming Soon" in the curriculum. I'd like to contribute content for this lesson.

Proposed lesson outline

The problem: Agents call tools autonomously. Today's audit trail is unsigned log files — editable, non-portable, and not independently verifiable.

The solution: Cryptographic receipts — Ed25519-signed proof of what an agent actually did, verifiable offline by anyone.

Practical walkthrough:

1. Wrap an MCP server with policy enforcement: npx protect-mcp -- node server.js
2. Define per-tool policies (allow/deny/rate-limit) using Cedar or JSON
3. Every tool call produces a signed receipt
4. Verify offline: npx @veritasacta/verify .protect-mcp-receipts.jsonl

Code samples: Three working examples at github.com/ScopeBlind/examples

Alignment with existing curriculum:

• Extends Lesson 6 (Building Trustworthy AI Agents) with cryptographic verification
• Extends Lesson 11 (Agentic Protocols — MCP) with security controls
• Complements the Agent Governance Toolkit, which already integrates protect-mcp receipts (PR #667)

Format: Happy to follow the existing lesson structure (Jupyter notebook + README + code samples) and submit as a PR.

npm: https://www.npmjs.com/package/protect-mcp (MIT)
IETF: https://datatracker.ietf.org/doc/draft-farley-acta-signed-receipts/

[github-actions]

👋 Thanks for contributing @tomjwxf! We will review the issue and get back to you soon.

[leestott]

Thanks @tomjwxf if you like to draft a contribution