Skip to content

Latest commit

 

History

10 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

CCS Integrations

Framework-native adapters for the Correctover Conformance Shape (CCS) runtime verification standard.

What is CCS?

CCS is a cryptographically verifiable runtime receipt standard for AI agents. Every tool call produces a signed, chain-linked receipt proving:

  • What was called — tool name, args digest, request hash
  • What happened — verdict (allow/block), response hash, behavior evidence
  • Who verified it — issuer identity, signing key fingerprint, deployment mode
  • When — timestamp, nonce, sequence number, expiry
  • Tamper evidence — Ed25519 signature over JCS-canonical JSON, previous-receipt hash chain

Receipts are 30-field L1 artifacts compatible with ccs-verifier==1.3.0, plus optional signed behavior evidence receipts.

Why integrations?

Instead of asking framework maintainers to build CCS support, we provide drop-in adapters. Install, configure, and your agent gets signed receipts — no framework fork, no PR to merge.

pip install ccs-pydantic-ai
from ccs_pydantic_ai import CCSToolset, CCSConfig

agent = Agent("openai:gpt-4o", toolsets=[CCSToolset(CCSConfig(deployment_mode="in-process"))])
result = await agent.run("Search for Q3 revenue")
# Receipts emitted automatically — signed, chain-linked, verifiable

Available Adapters

Framework Package Status Conformance
Pydantic AI ccs-pydantic-ai 🚧 In Development —
CrewAI ccs-crewai 📋 Planned —
AutoGen ccs-autogen 📋 Planned —
LangChain ccs-langchain 📋 Planned —
LlamaIndex ccs-llamaindex 📋 Planned —
MCP (generic) ccs-mcp 📋 Planned —

Architecture

┌─────────────────────────────────────────┐
│  Your Agent (Pydantic AI / CrewAI / …) │
└──────────────┬──────────────────────────┘
               │ tool call
               ▼
┌─────────────────────────────────────────┐
│  CCS Adapter (this repo)                │
│  ├── WrapperToolset / GuardrailProvider │
│  ├── Receipt Builder (30 fields)        │
│  ├── Ed25519 Signer                     │
│  └── Behavior Evidence Builder          │
└──────────────┬──────────────────────────┘
               │
        ┌──────┴──────┐
        ▼             ▼
   ┌─────────┐  ┌───────────┐
   │In-process│  │  Sidecar   │
   │ Key (seed)│  │  Key (ext) │
   └─────────┘  └───────────┘
               │
               ▼
        Signed Receipt (JSON)
        ├── L1 Receipt (authorization + chain integrity)
        └── Behavior Evidence (semantic verdict)
               │
               ▼
        ccs-verifier==1.3.0
        (strict parse + verify)

Two Deployment Modes

In-process Sidecar
Key storage Deterministic seed in app memory Private key outside agent process
Isolation Key shares process memory Key boundary enforced by OS
Use case Testing, dev, low-stakes Production, compliance, multi-tenant
Performance Native speed Local HTTP/loopback
Threat model Runtime integrity only Runtime + key exfiltration

Conformance Vectors

Reference test vectors are available at DSHCorrectover/ccs-conformance-vectors:

  • 10 signed vectors covering valid receipts, tamper detection, cross-key rejection, canonicalization, hash chain breaks, and replay
  • Two key models: sidecar (private key not in repo) and in-process (deterministic, reproducible)
  • Every L1 receipt is 30 fields, strict-compatible with ccs-verifier==1.3.0

Run verification:

git clone https://github.com/DSHCorrectover/ccs-conformance-vectors.git
cd ccs-conformance-vectors
pip install ccs-verifier==1.3.0 jcs cryptography
python3 verify_v131.py

Ecosystem Map

CCS interoperates with or is referenced by:

  • IETF SCITT — CCS as a reference receipt format for transparent auditing
  • IETF DMSC — CCS as application/runtime verification layer above DMSG
  • EMILIA — PR #668 merged, upstream tests pass with pinned ccs-verifier
  • x402 — Independent review identified 4 specification defects in their evidence binding
  • RootSign/PDR — Paired conformance vectors and field crosswalk in rootsign#37
  • AgenTrust — Third-party conformance run passed and merged

Specification

  • IETF draft: draft-correctover-ccs — an individual Internet-Draft; not an RFC or IETF endorsement, and the Datatracker page is authoritative for revision and status

Commercial Support

We help teams adopt CCS across their agent frameworks. Engagement is by email only — no calls.

  • Adapter development — building and hardening native CCS adapters for frameworks not yet covered
  • Production rollout — in-process and sidecar key modes, fail-closed enforcement, and signed-receipt pipelines
  • Conformance & audit — running the CCS conformance benchmark across your integrations with an Ed25519-signed audit trail

Scope and pricing are quoted per project after a short email scoping exchange. Contact: wangguigui@correctover.com.

License

  • Adapters in this repository: MIT
  • ccs-verifier core: ELv2 (pip-installable, closed-source)
  • Conformance vectors: CC0

Contributing

We need adapter maintainers for every major agent framework. If you use a framework that's not listed, open an issue or PR. Each adapter should:

  1. Use the framework's native interception point (not a fork)
  2. Produce 30-field L1 receipts strict-compatible with ccs-verifier==1.3.0
  3. Support both in-process and sidecar key modes
  4. Include tamper detection tests
  5. Pass the reference conformance vectors

About

Framework-native adapters for CCS (Correctover Conformance Shape) cryptographically signed runtime receipts — Ed25519 + JCS signed audit evidence for AI agent tool calls (Pydantic AI, CrewAI).

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages