Skip to content

dynamicfeed/df-verify

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 

Repository files navigation

DF-VERIFY/1 — verifiable grounding for AI that acts

Reference implementation of DF-VERIFY/1 — an open, vendor-neutral standard for cryptographically signing, publishing, and independently verifying exactly what an AI system was told the moment it acted.

When an AI acts — moves a robot, places a trade, files a claim — "trust me" is not an audit trail. DF-VERIFY attaches an Ed25519 signature to any JSON response, publishes the verifying key openly, and lets anyone check it with no account and no dependency on the issuer. You can verify, even against the issuer.

DF-VERIFY/1

What's here

Path What
clients/python dynamicfeed-verify — Python reference verifier (library + CLI)
clients/js @dynamicfeed/verify — JavaScript/TypeScript verifier (Node, Deno, Bun, browser)
clients/csharp C# reference verifier
examples/verified-agent a runnable agent that verifies a signature before it acts
tests/vectors language-agnostic conformance vectors + Python & JS harnesses

Verify in 30 seconds

Python

git clone https://github.com/dynamicfeed/df-verify && cd df-verify
pip install cryptography
python examples/verified-agent/agent.py            # verify a live verdict, then act
python examples/verified-agent/agent.py --tamper   # altered after signing → the agent refuses to act

JavaScript

import { verifyLive } from '@dynamicfeed/verify';
const { result } = await verifyLive();
if (!result.ok) throw new Error(`unverified world-state: ${result.error}`);

How it works

  1. Drop the signature block; keep the rest as the payload.
  2. Canonicalize — JSON, keys sorted recursively, compact separators, non-ASCII escaped \uXXXX, UTF-8.
  3. Fetch the public key from https://dynamicfeed.ai/.well-known/keys, look up signature.key_id.
  4. Verify the Ed25519 signature over the canonical bytes. Change one byte → it fails.

Conformance

python3 tests/verify_vectors.py     # Python harness  → "✓ ALL 10 VECTORS PASS"
node    tests/verify_vectors.mjs    # JS harness (run: npm install --prefix clients/js  first)

Both reference verifiers reproduce every vector — same canonical bytes, same signature verdicts — so you can confirm a new implementation in any language byte-for-byte.

Links

License

MIT.

About

DF-VERIFY/1 — verifiable grounding for AI that acts. Reference verifiers (Python/JS/C#), a demo agent, and conformance vectors. Spec: https://dynamicfeed.ai/standard

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors