Context: Agents will be one of the highest-risk parts of a Meshspaces rollout because wrong-target actions could post, fetch, or mutate in the wrong mesh. This issue is planning-only.
Goal: Define what agents, MCP surfaces, and automation layers will need in order to operate safely once Meshspaces exist.
Focus areas:
- per-mesh MCP target model vs shell-level MCP model
- explicit mesh targeting requirements for write actions
- agent onboarding and approval UX per meshspace
- quarantine and permission controls for agents
- wrong-target posting, fetch, and moderation risks
- how existing single-mesh agents might migrate without confusion
Expected deliverable:
- Open a PR from a Copilot branch with a report in
docs/meshspaces-review/agent-and-mcp-safety.md
- Include concrete API/UI contract recommendations and migration guardrails.
Ground rules:
- Do not build a meshspaces-aware MCP layer yet.
- Prefer planning docs, findings, and explicit interface proposals.
- Call out any current code assumptions that force single-instance behavior.
Context: Agents will be one of the highest-risk parts of a Meshspaces rollout because wrong-target actions could post, fetch, or mutate in the wrong mesh. This issue is planning-only.
Goal: Define what agents, MCP surfaces, and automation layers will need in order to operate safely once Meshspaces exist.
Focus areas:
Expected deliverable:
docs/meshspaces-review/agent-and-mcp-safety.mdGround rules: