Package
New package / cross-cutting
Problem Statement
pi-mono already exposes the right TypeScript SDK hooks for governance-oriented interception (tool_call, tool_result, and before_provider_request), but Agent Governance Toolkit does not currently have a first-class integration for it. That means pi-mono users cannot easily apply AgentMesh policy enforcement, bash safety review/deny rules, or tamper-evident audit logging in the same way other frameworks in the ecosystem can. In practice, this forces downstream projects to build and maintain custom wrappers instead of contributing a reusable upstream integration.
Proposed Solution
Add a new integration package under packages/agentmesh-integrations, for example pi-mono-agentmesh, with a published package such as @microsoft/agentmesh-pi-mono.
The package should:
- Attach AgentMesh policy evaluation to pi-mono SDK
tool_call hooks
- Record prompts, tool results, and provider requests in AgentMesh audit logs
- Support policy loading from YAML or JSON
- Add conservative bash safety review/deny checks for high-risk commands
- Expose a small helper for creating or wrapping a governed pi-mono SDK session
- Include TypeScript tests covering allow, deny, and review decisions plus audit/session behavior
- Include README/docs updates so pi-mono appears alongside the other framework integrations
This would let pi-mono users adopt AGT governance without custom sidecars or app-specific adapters.
Alternatives Considered
- Keep the integration downstream in application code. This works, but duplicates maintenance across projects and makes it harder for AGT users to discover or reuse.
- Add pi-mono support to the Python
agent-os adapter layer. That is a poorer fit because pi-mono's native integration surface is its TypeScript SDK and extension hooks.
- Use an external proxy or RPC wrapper instead of SDK hooks. That adds complexity and latency compared with direct in-process interception.
Priority
Important
Contribution
Package
New package / cross-cutting
Problem Statement
pi-mono already exposes the right TypeScript SDK hooks for governance-oriented interception (
tool_call,tool_result, andbefore_provider_request), but Agent Governance Toolkit does not currently have a first-class integration for it. That means pi-mono users cannot easily apply AgentMesh policy enforcement, bash safety review/deny rules, or tamper-evident audit logging in the same way other frameworks in the ecosystem can. In practice, this forces downstream projects to build and maintain custom wrappers instead of contributing a reusable upstream integration.Proposed Solution
Add a new integration package under
packages/agentmesh-integrations, for examplepi-mono-agentmesh, with a published package such as@microsoft/agentmesh-pi-mono.The package should:
tool_callhooksThis would let pi-mono users adopt AGT governance without custom sidecars or app-specific adapters.
Alternatives Considered
agent-osadapter layer. That is a poorer fit because pi-mono's native integration surface is its TypeScript SDK and extension hooks.Priority
Important
Contribution