Open-sourced one of my most commonly used harness practices.
Summary
Open-sourced a practice called FILETREE.md, which maintains a file with a complete file tree, AI-generated one-line descriptions, and change hashes for the project, serving as efficient context to help AI agents quickly understand the project structure.
View Cached Full Text
Cached at: 05/26/26, 03:12 PM
I’ve open-sourced one of my most commonly used harness practices.
In every project, I maintain a FILETREE.md: the full file tree, a one-sentence AI-generated description for each file, and a hash to detect changes.
Essentially, this pre-builds a file-level index for the project. For an agent, it’s high-density, high-efficiency context — especially in small to medium-sized projects. Adding this bit of information greatly helps the agent quickly grasp the entire project.
Similar Articles
@Xudong07452910: Strongly recommended open-source project: Trellis — an open-source harness that gives your AI coding agent team-level engineering standards. Trellis is the AI coding engineering framework I've been using, known as 'The best agent harness'. It…
Trellis is an open-source project that provides team-level engineering standards and a project memory framework for AI coding agents. It supports multiple platforms and enhances agent development efficiency and quality by persisting specifications and task context to code repositories.
@FakeMaidenMaker: awesome-harness-engineering — the knowledge in this project is far more valuable than the number suggests — it contains frontline engineering practices from OpenAI, Anthropic, Microsoft, and Meta. GitHub: https://github.com/ai-boos…
awesome-harness-engineering is a curated list of resources on AI agent harness engineering (context management, tool design, verification loops, memory systems, etc.) from companies like OpenAI, Anthropic, Microsoft, and Meta, aimed at helping developers build reliable agent frameworks.
@GitHub_Daily: When using AI-assisted programming, asking a simple question requires flipping through files one by one, which wastes tokens and easily leads to wrong context. codebase-memory-mcp parses the entire codebase into a knowledge graph, allowing AI to directly 'understand' the project structure. A single executable written in pure C, zero dependencies, …
codebase-memory-mcp is a tool written in pure C that parses the entire codebase into a knowledge graph, supports 158 programming languages, is compatible with 11 AI coding agent tools, greatly improving AI's understanding of project structure and reducing token consumption.
@Potatoloogs: When using Claude Code, Cursor, Codex to understand large projects, you often encounter a problem: every time you ask a question, it has to re-read files, find clues, and piece together context. Code is in src, docs in docs, design specs, screenshots, papers, videos scattered in other directories. Lots of material, but the relationships haven't been captured...
Graphify is a software engineering knowledge graph tool for AI coding assistants. It organizes project materials such as code, documents, and images into a queryable relationship graph, helping AI skip the step of repeatedly reading files when understanding large projects.
@wsl8297: If you have a bunch of PDFs, documents, project materials to feed to AI, Synthadoc is a direction worth looking at. GitHub: https://github.com/axoviq-ai/synthadoc… It compiles raw materials into a structured wiki at ingestion time, automatically...
Synthadoc is an open-source tool that compiles PDFs, documents, and other project materials into a structured local Markdown wiki, automatically establishing cross-references and detecting contradictions. It is suitable for personal or small teams for offline knowledge management.