Back to directory

DeepSeek Harness: From Power-On to Teardown

A first-hand, independently written guide to DeepSeek's open-source Harness agent framework, covering its system prompt, startup plugin list, real session logs, and a walk-through of using it after Claude Code or Codex.

DeepSeek Harness: From Power-On to Teardown

This is an independently written guide to DeepSeek Harness, the agent framework DeepSeek open-sourced under the MIT license, written by a non-programmer who installed it and tested it firsthand rather than paraphrasing the official release notes. It covers the exact system prompt sent to the model before each turn, a 129-line startup plugin manifest exported from the tool itself, three unedited raw session logs from its standard, PTC, and creation modes, a step-by-step account of the agent building itself a new tool, real billing figures from a PTC session, a list of extension packages available on npm but not part of the default installation, and a review of the project's commit history and incident postmortems.

It is distributed as free PDF, EPUB, and single-file HTML downloads with an accompanying short animation and slide deck, and is one entry in a series of similar guides to other agent tools and frameworks by the same author.

More in Books

See all 105
  • A long technical guide to AI agent memory: why stateless LLMs need it, the different memory architectures from vector stores to knowledge graphs, and how Cognee implements one as an open-source memory layer.

  • Samizdat

    rare-books.vercel.app

    A free, curated shelf of rare internal memos, leaked handbooks, and out-of-print books from technology and business history, collected for builders to read or borrow.

  • A free, web-only book that explains AI systems, from raw compute to a deployed and governed behavior, design-first, for a software engineer who knows distributed systems but is new to machine learning.