AIAny
AI Agent2026

dsh-context

Shows what fills a DeepSeek Harness agent's context and how it changes per request. Browse prompts, tools and messages, trace compaction, and inspect token usage, estimated costs and parent/subagent relationships.

Introduction

A context window can be nearly full even when the visible conversation looks short. Inspecting the assembled request reveals which tool schemas, injected instructions and tool results occupy that space, giving users a concrete starting point for diagnosing context growth.

What Sets It Apart
  • A request-level browser exposes actual prompt and message content, so large contributions can be traced to specific items.
  • Composition trends mark compaction and pruning, explaining sudden changes across steps and turns.
  • Cross-session usage views and an agent-family page connect individual conversations with broader activity.
  • File activity links operations back to tool results, tying context changes to the work that produced them.
Who It Suits and the Trade-offs

Useful for DSH users debugging long conversations, plugin authors inspecting prompt overhead, and teams tracking multi-agent work. Composition shares and costs include estimates; provider-reported totals are shown separately. Older steps reconstructed after compaction can be approximate, and host compatibility should be checked against the supported release matrix.

Information

  • Websitegithub.com
  • AuthorsBowen Liang
  • Published date2026/08/14

Categories

More Items

GitHub
AI Coding2026

Adds file editing, Git views, task monitoring and side chats to DeepSeek Harness. An extensible sidebar and split bottom workbench keep files, terminals and agent activity beside the conversation.

GitHub
AI Agent2026

Runs coding, research and document tasks through a desktop or Web agent. Its Cordis plugin system lets developers replace model adapters, tools and even the agent loop, or embed the runtime through SDKs.

GitHub

Autonomously generates, iteratively debugs, and composes robot control programs into a reusable skill library for discovering and refining manipulation skills across simulation and real robots. Uses a code-as-policy continual learning loop—accumulating validated sensorimotor skills instead of model weights—to tackle long-horizon household and benchmark tasks.