A full-stack AI red teaming platform by Tencent Zhuque Lab that scans AI agents, MCP servers, Agent Skills and AI infrastructure for security risks.
pic-standard/pic-standard
Open standard and Python verifier that require verifiable provenance and signed evidence before an AI agent executes high-impact tool calls.
- Stars
- 70
- Forks
- 48
- Open issues
- 20
- Last commit
- yesterday
What is it?
PIC (Provenance & Intent Contracts) is an open standard for checking AI agent actions before they run. An agent submits an action proposal that declares its intent, risk class, information sources, supporting claims, evidence, and the exact tool call. A verifier checks that structure and the attached evidence against keys the operator trusts, and the integration blocks execution when verification fails.
The main use is gating money movement, privacy-sensitive, and irreversible actions. Under PIC's core rule, those actions need at least one claim tied to trusted provenance, and self-asserted trust labels are sanitized to untrusted by default. Ed25519 signatures and SHA-256 hashes can be verified locally, and canonical attestations bind evidence to the tool, arguments, claims, and provenance IDs.
The Python package provides a CLI, an MCP server guard, an HTTP verification bridge with an OpenAPI contract, and a LangGraph node. A shared conformance suite is used alongside a separate TypeScript implementation. The project is at v0.9.1 and the specifications are still drafts ahead of v1.0.
Key features
- 01Verifies action proposals containing intent, impact class, provenance, claims, evidence, and the tool call.
- 02Requires trusted provenance for money, privacy, and irreversible actions.
- 03Verifies Ed25519 signatures against an operator-managed keyring with expiry and revocation support.
- 04Checks that live tool arguments match the verified proposal in the MCP guard and HTTP bridge.
- 05Runs locally without a hosted account or model API key.
Use cases
- Blocking an agent's payment or refund tool call unless a signed approval matches the exact amount and payee.
- Preventing an agent from deleting or exporting sensitive data based on an injected instruction.
- Adding a pre-execution verification gate to an MCP server or LangGraph workflow.
Quick start
- 1
Install the Python verifier with signature support
python -m pip install "pic-standard[crypto]" - 2
Clone the repository to run the sample fixtures
git clone https://github.com/pic-standard/pic-standard.git - 3
Install the cloned source in editable mode from the checkout
python -m pip install -e ".[crypto]"
Things to know
- The project is at version 0.9.1, so its API and specifications may still change.
- It requires you to configure trusted keys and enforce the decision in your own execution path.
- The LangGraph integration does not yet compare live arguments with the verified proposal.
In collections
Topics
FAQ
- What is pic-standard?
- Open standard and Python verifier that require verifiable provenance and signed evidence before an AI agent executes high-impact tool calls. PIC (Provenance & Intent Contracts) is an open standard for checking AI agent actions before they run. An agent submits an action proposal that declares its intent, risk class, information sources, supporting claims, evidence, and the exact tool call. A verifier checks that structure and the attached evidence against keys the operator trusts, and the integration blocks execution when verification fails.
- Is pic-standard free to use? What license does it have?
- pic-standard is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
- Is pic-standard still maintained?
- Actively maintained. The last commit to pic-standard was on Nov 6, 2026.
- What can I use pic-standard for?
- Blocking an agent's payment or refund tool call unless a signed approval matches the exact amount and payee. Preventing an agent from deleting or exporting sensitive data based on an injected instruction. Adding a pre-execution verification gate to an MCP server or LangGraph workflow.
- How do I install pic-standard?
- Install the Python verifier with signature support: python -m pip install "pic-standard[crypto]" Clone the repository to run the sample fixtures: git clone https://github.com/pic-standard/pic-standard.git Install the cloned source in editable mode from the checkout: python -m pip install -e ".[crypto]"
- Where can I get pic-standard?
- The source code and installation instructions are on GitHub: https://github.com/pic-standard/pic-standard
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
Open-source AI agent firewall that sits between agents and the network to block secret exfiltration, prompt injection, SSRF and MCP tool poisoning, with signed action receipts.
Open-source CLI and library that verifies AI model files, code, live LLM/agent/MCP endpoints, and recorded agent traces, reporting findings and gaps.
Access control for AI coding agents: blocks credential reads, holds risky actions for review, and logs every tool call across Claude Code, Codex, Cursor and MCP servers.
Security scanner for AI coding agents that checks generated code, MCP servers, prompts, and AI-suggested packages for risks.
Open-source authorization engine that checks every AI agent action and returns allow, deny, human_review, or compute before it runs.