Companion code for "Can Rust Make Unsafe AI Agent Actions Unrepresentable?"

This is a complete, runnable Cargo project. Its only job is to let you (a) confirm the example behaves as the article claims, and (b) capture the real compiler error for the article's "money shot" so it is authentic rather than hand-written.

Confirm the legal flow compiles and runs

cargo run

Expected output:

persisted key=refund_policy value="Refunds under $100 do not require manager approval." authority=Policy

That is the propose -> evaluate -> persist path working.

Capture the compiler error (the money shot)

1. Open src/main.rs.

2. At the bottom of main, uncomment the block marked // MONEY SHOT. It builds a fresh ProposedWrite and passes it straight to persist().

3. Run:

   cargo build
   

4. Copy the actual error[E0308] block rustc prints, verbatim, including the --> location line and the ^^^^ span underlines. Paste it into the article where the <!-- CAPTURE ... --> placeholder sits.

5. Re-comment the block so repo HEAD builds cleanly again.

Do not edit the compiler output for style. Its authenticity is the point; a Rust reader can tell hand-written rustc output at a glance.

Note

validate_authority accepts "security" and "policy" and rejects everything else, so you can also demonstrate the rejection path by changing the proposed authority to something unrecognized and running cargo run.