Feature: transfer verified crew memory between crews and across frameworks

I maintain cognicore-env — open-source (MIT) memory layer for agents, MCP server, no infrastructure required. Proposal plus offer of a PR.

CrewAI crews share memory inside a crew, which is the right primitive. The gap one level up: memory does not cross crews, and it never leaves CrewAI. A support crew that learned a company's actual refund policy (from resolved tickets) cannot hand that to a sales crew, or to a LangGraph or Claude Code agent, with any provenance.

Proposal: crew memory export/import as sealed bundles. Each memory travels with proof (command, result, exit code, output hash) and a chain of custody; importing crews receive memories in an observed state until locally re-verified; trust decays across unverified hops. After a public design review we added owner-per-concept (for conflicting facts, the owning crew's record wins on ties) and validity intervals (memory verified in March is not evidence about April) — these two decide most conflict and staleness questions in multi-team setups.

Use case: the support crew runs for a month and builds verified knowledge of how escalations actually resolve; the onboarding crew imports it as observed; new hires' questions get answers grounded in what really happened, with the proof trail attached.

Our transfer layer is implemented and tested across four platforms, and CrewAI's MCP support means this could ship as an integration without touching core. Would the team want an example, and should it sit in tools or docs?

Begin with CrewAI's MCP support and compare the proposed cognicore-env transfer layer across its four stated platforms. Review the design questions around sealed bundles, provenance, owner-per-concept, and validity intervals; a contribution would need a decided integration location (tools or docs), an example, and maintainer agreement.