dengyier

OpenWorkProof: Verify and Accept AI Agent Work

OpenWorkProof adds signed work contracts and portable evidence to AI agent workflows. It helps teams check authorization and execution independently, while keeping verification separate from a human’s decision to accept or reject the result.

NewPythonApache-2.0AI AgentsMCPSecurity
View on GitHub Compare
Stars
287
Forks
0
Stars, last 30 days
-26
Last commit
27 days ago
Contributors
1

Our take

Newless than six months old and changing fast

Health score

29/100

How it is scored
Activity
70
Community
4
Issues
21
Pull requests
20

-26 stars in the last 30 days

  • Fewer contributors than 91% of the projects we track

OpenWorkProof is a focused protocol for teams that need auditable agent work, but its rapid development and single-contributor footprint make independent evaluation important before relying on it in production.

Good fit if

  • You need portable evidence of an agent’s authorization, actions, and verification, rather than logs tied to one platform.
  • Your workflow can keep verification separate from a human or designated party’s acceptance decision.
  • You are prepared to evaluate a young Python project that is changing quickly and currently depends on one contributor.

Look elsewhere if

  • You need the framework to decide whether a result is correct or to accept work automatically on a customer’s behalf.
  • You need a mature project with demonstrated contribution from multiple maintainers and established issue-resolution history.
  • You are looking for payment, settlement, legal arbitration, or customer-adoption guarantees, which the project explicitly does not provide.
All health signals
Last commit2026-09-12 (27 days ago)
Commits, last 90 days100+
Releases, last 12 months5 (latest v1.4.0, 2026-08-30)
Contributors1
Issues closed, last 90 days0
Pull requests merged, last 90 days0
Project age2 months

Checked on 2026-10-10 with the GitHub API.

Project history

Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.

Selected snapshot

Stars

261

Forks

0

Change since Nov 18, 2026: 0 stars·0 forks

Overview

OpenWorkProof is a protocol and Python toolkit for recording what an AI agent was authorized to do, what it did, and what evidence supports the result. It addresses a gap left by agent communication protocols and ordinary logs: those can connect systems or record events, but do not by themselves establish that work was authorized, stayed within scope, or was accepted by the customer.

The project packages signed authorization, execution receipts, independent verification, and acceptance decisions into evidence that a third party can check offline. It is aimed at teams building agent platforms, CI workflows, or multi-agent systems that need reviewable delivery records. It does not determine whether every business outcome is correct or make the acceptance decision for a human.

Key Features

  • Models a work chain from a task and signed capability grant through policy decisions, action receipts, verification, and acceptance.
  • Uses Ed25519 signatures, normalized JSON, an append-only ledger, and content digests to support offline evidence checks.
  • Separates verification from customer acceptance, with distinct verifier and acceptor decisions.
  • Provides a Human Agency Profile for distinguishing delegated, reserved, and denied actions before execution.
  • Offers CLI, Python API, MCP, GitHub Action, and AgentTeams integration paths.
  • Builds and verifies Surface and Acceptance bundles for portable review outside the original system.
  • Supports checks that the verification process covered its agreed population and that negative controls failed as expected.

Use Cases

  • A software team can require signed authorization and evidence when an agent edits a repository or runs tests, then let a separate reviewer assess the result.
  • A provider delivering agent-generated work can give a customer an offline evidence bundle instead of relying only on logs in the provider’s platform.
  • An organization using multiple agents can separate execution, verification, and acceptance roles to avoid one agent approving its own work.
  • A team embedding agents in MCP or CI workflows can check whether tool use falls within task-specific permissions before an action runs.

What you need

Detected in the repository

  • Python >=3.10 (from pyproject.toml)
  • A Dockerfile, so it can run in a container
  • A test suite and automated checks on GitHub Actions

License in plain words

Apache-2.0permissive

  • Commercial use: yes
  • Modify and redistribute: yes
  • You must keep: the license, the NOTICE file and a note of your changes
  • Share your changes: no
  • Includes an explicit patent grant from the contributors:

A summary, not legal advice: the LICENSE file is what applies.

Getting Started

Install the published Python package and inspect the CLI:

python -m pip install openworkproof
owp --help

For source installation, examples, and integration details, see the repository README.

Alternatives

  • agent-governance-toolkit: It enforces agent policies and logs tool-call decisions, rather than issuing signed work contracts and portable execution evidence.
ProjectLanguageLicenseStarsStatus
OpenWorkProofPythonApache-2.0287New
agent-governance-toolkitPythonMIT6.6kActive

Considerations

The project is new and changing quickly. Recent activity and releases show active development, but the measured contributor data shows that work is concentrated in one person. The measured period also shows no issues closed or pull requests merged, so there is little evidence yet of a broader review and maintenance process.

Treat the protocol’s conclusions narrowly: verification means evidence supports a conclusion under the configured scope, not that a business result is universally correct. Acceptance, payment, legal recognition, and adoption remain separate external facts. The README also describes the DeepSeek Harness plugin as a local candidate rather than a published npm release, so do not assume that integration is available as a released package.

Found this useful?

Share it with someone who would like OpenWorkProof.

Comparisons