kv-cache 6

DeepSeek Harness plugin: DietrichGebert/ponytail lazy senior mode & 7-rung ladder port, 6 skills with bilingual descriptions & per-skill toggles, zero tools, zero-cache-miss | DeepSeek Harness 插件：DietrichGebert/ponytail 懒人 senior 模式与七阶梯子完美移植，6 个技能描述双语自由切换

Context compression tool (context_compact): the agent writes the replacement checkpoint itself and hands it to the host compaction engine, which skips the LLM summarizer call; automatic compaction stays on the official engine.

Deterministic system-prompt section ordering for DeepSeek Harness — fixes KV-cache invalidation from unstable same-order section ordering across session resume. / 确定性 system-prompt section 排序：修复会话恢复时同 order section 排序漂移导致的 KV 缓存断层。

DeepSeek Harness plugin: attaches a stable per-conversation session header to model requests routed to configured providers. Header name is configurable, so one plugin serves SMG (X-SMG-Routing-Key), OpenCode (x-opencode-session), and any other header-bas

Fold the DSH skill catalog prompt surface: static KV-cache-stable catalog + BM25/bge-m3 hybrid skill_search + autoRoute hints. Rendering-only fold: source.entries stay untouched, the skill tool and /name injection stay host-owned. v0.3.0 adds a semantic r

Persistent KV-cache/session-state manager for DeepSeek Harness (llama.cpp slot snapshots)