04 / Work
Implementations and qualifications.
Some Yukh work lives in current repositories; some is now evidence-driven qualification of external machinery. This page separates the two so research candidates are never mistaken for adopted components.
Current Yukh implementations
Foundation / current capability-policy implementation
Yukh MCP
Current directionPreserve the capability decision and evidence boundary while proving that Yukh policy composes with host-native security instead of duplicating or bypassing it. Durable production readiness remains unproven.
Understand the responsibility View Yukh MCP repository
Published / synthetically qualified GitHub adapter
Yukh Projects
Current directionKeep the proven GitHub Projects implementation useful while re-evaluating its architectural role as an adapter for abstract accepted work state. Publication and synthetic qualification do not imply a live apply qualification.
Understand the responsibility View Yukh Projects repository
Preparation evidence complete / current coordination kernel
Yukh Coordination
Current directionPreserve claims, conflict, handoff, attribution and replay semantics while allowing transports, workspaces and human interaction surfaces to remain replaceable. Public/live runtime qualification remains unproven.
Understand the responsibility View Yukh Coordination repositoryActive qualification work
These are investigations, not additions to the Yukh suite. A candidate may pass, fail, remain optional, or prove that Yukh should own less code.
| Concern | Candidate(s) | Current evidence / next gate | Governing record |
|---|---|---|---|
| Track A / Agent host | goose + Hermes | Runtime-substrate PASS for both. A2 Yukh/host two-sided policy composition is defined and awaits executable candidate adapters. | Issue |
| Shared memory / Track C2 | TencentDB Agent Memory | Qualify team memory as contextual machinery that cannot become Yukh authority, accepted state or evidence truth. | Issue |
| Durable execution | Apache Maka | Qualify durable AgentRun and execution/evidence facts without importing private reasoning or runtime authority. | Issue |
| Orchestration | OpenHuman | Evaluate checkpointed workflows and approvals as replaceable machinery, independently from shared memory. | Issue |
| Execution isolation | OpenHands | Test whether bounded Yukh capabilities can execute through an external sandbox and return correlated evidence. | Issue |
| Workspace | Buzz | Evaluate human-agent collaboration as a replaceable surface over Yukh coordination semantics. | Issue |