VCC-style instant, near-lossless deterministic compaction engine for the DeepSeek Harness — a drop-in replacement for @deepseek-ai/dsh-compaction-basic

Zero-LLM compaction engine for DSH: folds from the model-maintained checkpoint file or a mechanical span digest; LLM replay is the dead-last fallback.

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.

A framework-agnostic budget tracker and tail-preserving compaction engine for agent conversation history.

Headroom context-compression backend for DeepSeek Harness: local proxy lifecycle, a compaction engine backend, the headroom_retrieve CCR tool, and a settings card.

Lunar Proactive AI Skill System & Hierarchical 10-in-1 Compaction Engine for AI Agents & CLI Workflows

Token-meter-driven compaction policy and LLM summarization backend for the DeepSeek Harness

Abstract compaction service seam (ctx.compaction) for the DeepSeek Harness

Replay-safe model-free head/middle/tail pruning for tool-result surface nodes

Human-facing slash command for explicit session compaction

Framework-agnostic layout-engine adapter for snapgrid (thin layer over react-grid-layout/core).

Non-destructive, provider-independent context management for Pi.

The OpenCode harness connects `HarnessAgent` to OpenCode through a sandboxed bridge.

Conversation domain: skeleton, ordered chat flow, composer with the Host-backed busy-Enter preference, and details host

Engine for Shiki using Oniguruma RegExp engine in WebAssembly

Engine for Shiki using JavaScript's native RegExp

Framework-agnostic dnd-kit grid engine for snapgrid — the shared drag/resize/cross-grid brain consumed by @snapgridjs/react (and future Vue/Solid bindings).

Expose context-mode MCP tools as native DeepSeek Harness tools

styled() API wrapper package for emotion.

This package is intended for Prisma's internal use