obsigna

KO | EN

gitlite — search

#python #react #ai #typescript #go #golang #linux #docker #macos #awesome-list #pytorch #ios #mcp #open-source

(★ 20)

Agent Receipts — cryptographically signed audit trails for AI agent actions. Protocol spec, SDKs (Go, TypeScript, Python), and MCP proxy.

obsigna 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Agent Receipts

**Cryptographically signed audit trails for AI agent actions**

[![Go Tests](https://github.com/agent-receipts/obsigna/actions/workflows/sdk-go.yml/badge.svg)](https://github.com/agent-receipts/obsigna/actions/workflows/sdk-go.yml)
[![TS Tests](https://github.com/agent-receipts/obsigna/actions/workflows/sdk-ts.yml/badge.svg)](https://github.com/agent-receipts/obsigna/actions/workflows/sdk-ts.yml)
[![Python Tests](https://github.com/agent-receipts/obsigna/actions/workflows/sdk-py.yml/badge.svg)](https://github.com/agent-receipts/obsigna/actions/workflows/sdk-py.yml)
[![Daemon](https://github.com/agent-receipts/obsigna/actions/workflows/daemon.yml/badge.svg)](https://github.com/agent-receipts/obsigna/actions/workflows/daemon.yml)
[![MCP Proxy](https://github.com/agent-receipts/obsigna/actions/workflows/mcp-proxy.yml/badge.svg)](https://github.com/agent-receipts/obsigna/actions/workflows/mcp-proxy.yml)
[![Hook](https://github.com/agent-receipts/obsigna/actions/workflows/hook.yml/badge.svg)](https://github.com/agent-receipts/obsigna/actions/workflows/hook.yml)
[![License: Apache-2.0](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://raw.githubusercontent.com/agent-receipts/obsigna/main/LICENSE)

</div>

| | |
|---|---|
| **Protocol site & spec** | [agentreceipts.ai](https://agentreceipts.ai) |
| **Conformance** | [Cross-language interop suite](https://agentreceipts.ai/conformance/) |
| **Tooling docs** | [obsigna.dev](https://obsigna.dev) |
| **Daemon setup & migration guide** | [obsigna.dev/getting-started/daemon-setup/](https://obsigna.dev/getting-started/daemon-setup/) |
| **API reference** | [Go](https://obsigna.dev/sdk-go/api-reference/) · [TypeScript](https://obsigna.dev/sdk-ts/api-reference/) · [Python](https://obsigna.dev/sdk-py/api-reference/) |
| **Examples** | [obsigna-examples](https://github.com/agent-receipts/obsigna-examples) |
| **Blog** | [Your AI Agent Just Sent an Email](https://jongerius.solutions/post/your-ai-agent-just-sent-an-email/) · [Every MCP Tool Call My AI Makes Now Gets a Signed Receipt](https://jongerius.solutions/post/auditing-github-mcp-agent-receipts/) |
| **Go** | [sdk/go](https://pkg.go.dev/obsigna.dev/sdk/go) · [mcp-proxy](https://pkg.go.dev/github.com/agent-receipts/ar/mcp-proxy) · [dashboard](https://pkg.go.dev/github.com/agent-receipts/dashboard) |
| **npm** | [@obsigna/sdk-ts](https://www.npmjs.com/package/@obsigna/sdk-ts) |
| **PyPI** | [obsigna](https://pypi.org/project/obsigna/) |

---

## What is this?

**Agent Receipts** is an open protocol for producing cryptographically signed, tamper-evident records of AI agent actions. It defines the receipt format, signing scheme, chain structure, and taxonomy of action types. Anyone can implement it — in any language, in any runtime.

**Obsigna** is the reference toolset that implements the protocol:

| Tool | What it does |
|------|-------------|
| `obsigna-mcp` | MCP stdio proxy — signs every tool call, adds policy hooks |
| `obsigna-daemon` | Out-of-process signing daemon — holds the key, owns the audit chain |
| `obsigna-hook` | PostToolUse hook for Claude Code and other runtimes |
| `obsigna` | CLI for browsing and verifying receipt databases |
| `sdk/go`, `@obsigna/sdk-ts`, `obsigna` (Python) | SDKs for embedding receipt creation in your own code |

## Conformance

Anyone can implement the protocol — so interoperability is the property that matters. The three SDKs (Go, Python, TypeScript) are separate, no-shared-code implementations that **verify one another's receipts**: every conformance vector is signed once with a shared Ed25519 keypair, and each SDK re-verifies the signatures, canonical JSON, and hashes the other two produced — across a positive corpus, a MUST-reject corpus, and four pinned spec versions. This cross-language verification is enforced in CI (it catches serialisation drift between the languages, not spec-interpretation ambiguity — the SDKs share one author).

The **[Conformance page](https://agentreceipts.ai/conformance/)** is the citeable summary: the interop claim, a results matrix with per-set vector counts, permanent links to every frozen vector set, and how to reproduce the checks. The shared corpus lives in [`cross-sdk-tests/`](cross-sdk-tests/).

## Formal verification

Where conformance shows the implementations agree, formal verification shows the **protocol logic itself is sound**. The receipt chain's tamper-evidence property (spec §7) is machine-checked with a bounded [Alloy](https://alloytools.org/) model in [`formal/chain-invariants/`](formal/chain-invariants/): over all instances up to the checked scopes, any receipt sequence that passes §7.3 verification is a genuine chain the issuer signed, in issuer order — so no modification, insertion, interior deletion, reorder, or cross-chain splice by a party **without the issuer's key** can produce a different sequence that still verifies. The one residual, tail truncation of an open chain, is the documented floor of §7.3.1, reproduced as an explicit fact rather than hidden.

This is bounded model checking **relative to the cryptographic assumptions** (collision-resistant hashing, unforgeable signatures) — it does not prove the primitives, and the Byzantine key-holder case is named as the boundary, not covered. Run it yourself with [`formal/chain-invariants/run.sh`](formal/chain-invariants/run.sh); the decision, threat model, scope, and honest statement of guarantee are in [ADR-0039](docs/adr/0039-formal-verification-chain-invariants.md).

## Start here

Both paths below require the daemon — it holds the signing key and owns the audit chain. Install it first:

```bash
brew install agent-receipts/tap/obsigna
obsigna daemon start
```

**Fastest path — PostToolUse hook (Claude Code):** one config snippet and every tool call gets a signed receipt automatically:

```json
{
  "hooks": {
    "PostToolUse": [{ "matcher": "", "hooks": [{ "type": "command", "command": "obsigna-hook" }] }]
  }
}
```

Add that to `~/.claude/settings.json`, then inspect the audit trail:

```bash
obsigna list
obsigna show <seq>
obsigna verify
```

**More control — MCP proxy:** wraps any MCP server and adds policy hooks and risk scoring on top of signed receipts:

- [Claude Desktop setup](https://obsigna.dev/mcp-proxy/claude-desktop/)
- [Claude Code setup](https://obsigna.dev/mcp-proxy/claude-code/)
- [Codex setup](https://obsigna.dev/mcp-proxy/codex/)

## Project layout

| Project | Description |
|---------|-------------|
| [`spec/`](spec/) | Protocol specification, JSON schemas, governance |
| [`sdk/go/`](sdk/go/) | Go SDK |
| [`sdk/ts/`](sdk/ts/) | TypeScript SDK |
| [`sdk/py/`](sdk/py/) | Python SDK |
| [`daemon/`](daemon/) | Signing daemon — out-of-process key custody, shared audit chain |
| [`mcp-proxy/`](mcp-proxy/) | MCP proxy with receipt signing, policy engine, intent tracking |
| [`hook/`](hook/) | PostToolUse hook binary for Claude Code and other runtimes |
| [`cross-sdk-tests/`](cross-sdk-tests/) | Cross-language verification tests |
| [`formal/`](formal/) | Formal models (Alloy) machine-checking protocol properties |
| [`docs/adr/`](docs/adr/) | Architecture Decision Records |
| [dashboard](https://github.com/agent-receipts/dashboard) | Local web UI for browsing and verifying receipt databases |
| [openclaw](https://github.com/agent-receipts/openclaw) | Agent Receipts plugin for OpenClaw |
| [obsigna-examples](https://github.com/agent-receipts/obsigna-examples) | Worked examples and integration demos |

## SDK quick start

> **Choose your trust model.** The snippets below keep the signing key inside the
> agent process — a deliberate deployment model where the agent host is trusted
> and tamper-evidence is aimed at downstream parties. In this model, anyone with
> code execution in the agent can forge receipts. To defend against a compromised
> agent, use the
> [daemon-mediated path](https://obsigna.dev/getting-started/daemon-setup/),
> where a separate daemon owns the key and your app only sends events over a
> socket. See the [Trust Model](https://agentreceipts.ai/specification/trust-model/)
> page for the full spectrum (in-process → daemon-isolated → HSM/KMS).

### Go

```bash
go get obsigna.dev/sdk/go
```

```go
import "obsigna.dev/sdk/go/receipt"

keys, _ := receipt.GenerateKeyPair()
unsigned := receipt.Create(receipt.CreateInput{
    Issuer:    receipt.Issuer{ID: "did:agent:my-agent"},
    Principal: receipt.Principal{ID: "did:user:alice"},
    Action:    receipt.Action{Type: "filesystem.file.read", RiskLevel: receipt.RiskLow},
    Outcome:   receipt.Outcome{Status: receipt.StatusSuccess},
    Chain:     receipt.Chain{Sequence: 1, ChainID: "chain_1"},
})
signed, _ := receipt.Sign(unsigned, keys.PrivateKey, "did:agent:my-agent#key-1")
```

### TypeScript

```bash
npm install @obsigna/sdk-ts
```

```typescript
import {
  createReceipt,
  generateKeyPair,
  signReceipt,
} from "@obsigna/sdk-ts";

### Python

```bash
pip install obsigna
```

```python
from obsigna import (
    create_receipt, generate_key_pair, sign_receipt,
    CreateReceiptInput, Issuer, Principal, Outcome, Chain,
)
from obsigna.receipt.create import ActionInput

keys = generate_key_pair()
unsigned = create_receipt(CreateReceiptInput(
    issuer=Issuer(id="did:agent:my-agent"),
    principal=Principal(id="did:user:alice"),
    action=ActionInput(type="filesystem.file.read", risk_level="low"),
    outcome=Outcome(status="success"),
    chain=Chain(sequence=1, previous_receipt_hash=None, chain_id="chain_1"),
))
signed = sign_receipt(unsigned, keys.private_key, "did:agent:my-agent#key-1")
```

See the [Python SDK README](sdk/py/README.md) for the full quick start and daemon delivery.

## Contributing

See [CONTRIBUTING.md](CONTRIBUTING.md) for development setup and PR guidelines.

## Security

See [SECURITY.md](SECURITY.md) to report vulnerabilities. The [threat model](docs/threat-model.md) documents trust boundaries, in-scope and out-of-scope threats, and the mitigation roadmap.

## License

Apache License 2.0 -- see [LICENSE](LICENSE).
The protocol specification in `spec/` is licensed under MIT.

obsigna

obsigna 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

obsigna

obsigna 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README