Blog
DeepSeek Harness Developer Preview: Everything Is a Plugin
DeepSeek Harness Developer Preview (v0.1) is now open for testing under the MIT license: models, tools, skills, sessions, sandboxes, loops, scheduling, and UI are all composed from plugins. The four run modes and how to get started.
OpenClaw: Core Architecture, How It Works, and Agent Deployment
OpenClaw's core architecture, how it works, and multi-agent deployment: workspace bootstrap files, per-session transient agents, session compaction/pruning memory, and the sessions_send vs sessions_spawn config.
DeepSeek Harness Hands-On: What the Other Half, Beyond the Model, Actually Buys You
A hands-on record of DeepSeek Harness after its open-source release: getting all four entry points (Web/Headless/Python SDK) running, whether Trajectory and the system prompt can be reconstructed from logs, a same-model (Kimi K3) tool-trajectory comparison against Kimi Code, and the same V4 Pro building a jump game under each harness.
Inside DeepSeek Harness: An Agent Architecture Built Entirely from Plugins
A teardown of DeepSeek Harness's plugin-based runtime: Cordis's Fiber lifecycle and reversible effects, the preset two-layer scope chain, Code Mode's worker_threads isolation, and the layered tool-shadowing algorithm — compared point by point against Codex's engineering choices.
Taming AI Coding: A Team Playbook for Harness Engineering
From the six pillars of Harness Engineering to concrete MCP, Skills, knowledge-base, and multi-agent setups inside CodeBuddy, through a three-phase rollout roadmap and a team-built harness-audit compliance-check skill — a team-level AI coding governance playbook you can follow step by step.
Is Loop Engineering Dead? A Guide to Graph Engineering
From Loop Engineering to Graph Engineering: the five-layer evolution, Loop's five structural flaws, the four-part graph model (nodes/edges/state/policy), three classic orchestration topologies, and real production case studies from LangGraph, Uber, and LinkedIn.
One person, one team: a minimal dev workflow with OpenSpec + Superpowers
Two open-source tools together: OpenSpec locks down your design intent, Superpowers runs TDD execution. A single architect can ship a tested kanban system from scratch in 30 minutes. With full commands and code.
Superpowers + gstack in Practice: 2 Plugins, 37 Skills, 5 Handoff Points, and a Complete Dev Loop
A deep dive into how Superpowers and gstack complement each other: Superpowers enforces code quality (TDD, debugging, review), gstack owns the product lifecycle (strategy, QA, release). With source-level analysis, 5 key handoff points, two integration patterns, and a CLAUDE.md template, this guide shows how to build a complete development loop from idea to production.
GStack in Practice: How AI Tooling Built This Blog From Scratch
GStack is a skill system running on Claude Code that covers the full development lifecycle — from product design to code review. Here's how I used it to design, build, and test this blog in half a day.
Superpowers in Action: How AI Coding Turned Two Weeks of Work Into Half a Day
A real-world account of using Claude Code to go from a vague idea to a fully deployed blog — product design, code review, testing, and deployment — in half a day. The same work would take two weeks without AI.