@deepseek-rlm/dsh-rlm-jupyter

The concrete ctx.rlm provider. It owns one lazy persistent IPython kernel per exact live DSH Agent, authenticated Jupyter v5 shell/IOPub/control channels, FIFO cell execution, generation-fenced host.request comms, managed Python 3.11 provisioning, snapshot/restore, interruption, and process-tree cleanup.

The host bridge translates Prime-compatible requests into ctx.llm, ctx.subagents, and ctx.tools; it never calls a provider or drives an agent loop itself. Full rc.7 operation requires the ordered compatibility patches documented at the repository root.

When rlm(..., model=...) omits its model selector, the child inherits the exact provider/model from the parent session's current request/header, including Agent-scoped Web model selection. Only a direct pre-request bridge call with no request header falls back to the parent's construction options. An explicit selector still chooses its named route, and omitting thinking does not implicitly copy a reasoning effort.

Construction validates that subagentProvider resolves to an exact native DSH provider and fails loudly when it does not. The installable bundle gates this constructor with its rlmSpawnReady hard dependency; Loader row order is not a readiness contract. Standalone compositions must likewise make their configured provider available before activating this package.

Artifacts are isolated under /sessions/. Historical cells are never replayed. Direct Python and %%bash execution has the kernel process’s OS authority and is not constrained by DSH tool policy.

See the root README.md for configuration, installation, security, and troubleshooting.