R2

Runtime-local LLM route capability declarations for DeepSeek Harness integrations.

A thin Feishu/Lark WebSocket adapter for the durable DSH assistant-delivery core.

Owner-authorized GitHub action facade with embedded compatibility and an external Unix broker.

Capture independently accepted tool workflows and replay versioned skills through the current native Goal and tool authority.

A bounded active-hours heartbeat policy layer backed by assistant-automations.

Persistent owner-scoped business goals over the native DSH goal domain.

Owner-authorized goal event opportunity filtering with durable utility decisions and cooldowns.

Scoped, bounded, approval-gated long-term memory for DeepSeek Harness agents.

Approval-gated behavioural self-evolution for the DSH personal assistant.

A DeepSeek Harness plugin for Assistant Policy.

A DeepSeek Harness plugin for Assistant Deepseek Budget.

A core-only meta-bundle for the DSH personal assistant policy, memory, Wiki, and automations plugins.

Durable supervised workflow learning with approval, replay, effect-blocked shadow, one-shot canary, and exact rollback.

Finite owner-authorized offline Linux Docker jobs with durable audit and external revocation.

A Super Relay OpenAI Responses route with conservative goal-budget token metering.

Bounded growth agent with durable reviews driven by owner-verified task feedback and frozen source models.

A safe, searchable, approval-gated Markdown knowledge vault for DeepSeek Harness.

A bounded, append-only outcome evaluation ledger for the DSH personal assistant.

A durable, policy-gated inbound/outbound message core for DeepSeek Harness.

Durable ordinary-task memory learning through native Automations, independent review and bounded adoption.

Policy-gated, content-free health and readiness aggregation for DSH assistant services.

Bounded, typed preference learning for the DSH personal assistant.

A durable, fenced cold-start scheduler and isolated Agent runner for DeepSeek Harness.