Skill Port
Store a Skill once. Skill Port uses a local Hub, explicit enablement, and source records, then serves multiple projects and agents through CLI and Desktop clients.
Content index
Developer tools, agent workflows, and engineering practice.
15 entries
Work
05 PROJECTS2026
QuotaBar
A macOS menu-bar utility for several AI coding services. It surfaces the tightest available metric and lets one provider fail without blocking the others.
Kane Skills (Chinese original)
A public, self-contained Agent Skills library that packages content production, engineering governance, skill evolution, visual creation, and delivery workflows into reusable skills.
Memories Are Diamonds (Chinese original)
A journal-style agent that uses Markdown and a TypeScript CLI to preserve real life, with AI helping organize, ask follow-up questions, and review while the person keeps final judgment.
Report Archive (Chinese original)
A public, chronological index for standalone research reports and interactive pages, deployed automatically through GitHub Pages.
Writing
10 ARTICLESLearning Pi Through Practice (Chinese original)
A practice-first path for learning Pi: start with real work, add only repeated capabilities, and tune context after the workflow becomes familiar.
How an Agent Recovers a Project Without Chat History
Turn a repository into an agent workspace so a fresh session can recover project facts, state, and the correct workflow from files alone.
Hard Gates for Fast Agents
Use observable acceptance criteria, risk-based tests, hard gates, and independent review to keep a fast Coding Agent on the intended path.
Parallel Work Does Not Save Tokens (Chinese original)
Break down the full agent cost chain, separate time gains from token billing, and choose tools and multi-agent collaboration by measurable cost.
Where Coding Agent Tokens Go (Chinese original)
A practical accounting of system prompts, repository context, tools, retries, and model routing shows where coding-agent costs really accumulate.
The AI-Native Development System
Use the repository to hold facts, process, state, and experience, then control agent risk with runnable slices, hard gates, and feedback loops.
AI-Native Development: From Design to a Running Version (Chinese original)
Use executable prototypes, small runnable releases, stage gates, and observable feedback to constrain a fast coding agent.
AI Coding Agent Landscape: From Tool Selection to Architecture
A layered comparison of eight AI coding agents, covering architecture, engineering depth, enterprise requirements, and field-delivery scenarios.
Why the Ministry-Style Agent Pipeline Fails (Chinese original)
Role boundaries and handoffs can make a multi-agent system legible, but context loss, semantic compression, and coordination cost often outweigh the apparent structure.
From Notes to an LLM-Driven Structured Wiki (Chinese original)
Move Obsidian notes into a structured wiki maintained by an LLM, with raw sources, a knowledge network, and explicit operating rules.
No matching content
Try another keyword, or clear the filters to see everything.