Blocklace A2A

Cryptographic audit layer for the A2A protocol.

Problem

A2A secures the transport layer (TLS, OAuth). It does not secure the message layer:

• No audit trail — messages aren't hash-chained
• No tamper detection — records can be modified after delivery
• No equivocation detection — an agent can send conflicting messages to different parties

Solution

Blocklace adds message-level cryptographic guarantees on top of A2A:

Blocklace complements A2A. It does not replace TLS/OAuth.

How It Works

Each message becomes a signed block in a DAG:

Block {
  author: "org-a"
  content: "Approved: $100"
  parents: ["67a3e7bf"]      ← hash pointers to prior blocks
  hash: sha256(above)
  signature: ed25519(hash)
}

Equivocation is detected when the same author creates two blocks where neither is an ancestor of the other.

Demo

$ python demo.py

Registering agents...
  [OK] org-a (pk: 4a4ccb1f...)
  [OK] org-b (pk: 3aff0ee5...)
  [OK] org-c (pk: 736199f4...)

Appending blocks...
  [84c7b686] author=org-a parents=[]         content="Hello from A"
  [44ed8588] author=org-b parents=[84c7b686] content="Hello from B"
  [67a3e7bf] author=org-a parents=[44ed8588] content="Reply from A"

Simulating equivocation (org-c sends conflicting messages)...
  [63eb9c6b] author=org-c parents=[67a3e7bf] content="Approved: $100"
  [35331451] author=org-c parents=[67a3e7bf] content="Approved: $999"

Equivocation detected:
  author:    org-c
  block_1:   63eb9c6b (content="Approved: $100")
  block_2:   35331451 (content="Approved: $999")
  evidence:  Blocks share parent [67a3e7bf] with no causal relationship

Installation

pip install git+https://github.com/en-yao/blocklace-a2a.git

Requires Python 3.10+.

Quick Start

from blocklace_a2a import Blocklace, AgentId

lace = Blocklace()
keys = lace.register_agent(AgentId("org-a"))
result = lace.append(keys, "Hello from A")
print(result.block.short_hash)  # "84c7b686"

Integration

Wrap A2A message passing with middleware:

from blocklace_a2a import Blocklace, create_middleware

lace = Blocklace()
middleware = create_middleware(lace, "my-agent")

# Outgoing
envelope = middleware.wrap_outgoing({"task": "process", "data": "..."})
payload = envelope.to_dict()  # JSON-serializable

# Incoming
result = middleware.verify_incoming(received_envelope)
if not result.valid:
    raise SecurityError(result.errors)

See examples/a2a_integration.py for a complete example.

Limitations

Status

Proof of concept. Suitable for experimentation and learning. Not production-ready.

References

• A2A Protocol
• The Blocklace: A Byzantine-repelling and Universal CRDT (Almeida & Shapiro, 2024)

License

MIT