This plugin was developed using DeepSeek Harness and the DeepSeek V4 Flash 0731 / Pro 0813 models, with multiple rounds of refinement. It enhances the existing sub‑agent functionality and adds several other features.
Delete one finished turn from a DSH session without deleting the session. A trash button removes that turn's question, reply and tool records from the conversation and the model context, while the session, later turns and the append-only log stay. Independent continuation of hanshenmesen/dsh-turn-delete (DSH 0.1.5-rc.2 + 0.2.0-rc.1).
DSH (DeepSeek Harness) mobile remote mode plugin - LAN discovery, QR pairing, Level-1 observer and Level-2 remote prompt from Android/iOS phones and any browser. Multi-platform, early development (Dev0.1).
Assembled context for DeepSeek Harness: a context tree over the session surface with per-node assemble modes (full / key / off), agent-authored presets, and a panel that decides what the model actually sees.
Context compression for DSH: visually configure summary and recent-context budgets (8k + 16k by default), keep tasks on track, and search full conversation history. Friendly to small-context models.
Solidify DSH session history step by step: rewrite tool results in place to the facts the model actually kept, so long conversations stop re-sending noise every turn. Costs tokens and time: a ~770-character contract in every request, plus one extra full-context request and round-trip per step when stepSummary is on.
Persistent, model-editable memory/notes store for DeepSeek Harness. Adds memory_set/get/delete/search tools backed by ctx.storageDomain so facts survive across sessions (dsh-plugin).
System-1 decision models (Jev/Laya/Kev-class) as the governance layer for an LLM agent context lifecycle: growing association graph over session segments, relevance-gated recall with Trace-as-State ordering, and plan pre-ranking — measured in solve rate, cache hit/miss tokens, cost and latency. DSH plugin + harness-agnostic proxy.
DSH plugin: hands a long session over to a fresh session in the same workspace on /handoff or at a token threshold, carrying the preset, model, permissions and plan mode across; the brief is written by the session's own model and the switch waits for a step boundary instead of interrupting running work.
