AI agent deep dives
Plan mode, hooks, subagents, MCP — get the agent to actually ship.
10 principles for high-quality agent system prompts
Not the "act as an expert" cliché. These are lessons from agents that actually ship in production.
Read more →Wire MCP servers into the agent: Linear / Notion / GitHub
One config block; the agent immediately gains every resource and tool the MCP server exposes.
Read more →Subagent collaboration: parallel work patterns
The main agent spawns up to 4 concurrent subagents; depth ≤ 2 — balanced parallelism without runaway recursion.
Read more →Agent hooks 101: PreToolUse / PostToolUse
Six lifecycle points, each can run a shell script: audit, cache, block, rewrite.
Read more →Browser agent in practice: scraping Xiaohongshu into Kition
Xiaohongshu has no public API. Kition’s real Chromium browser agent logs in, scrapes content, and files it.
Read more →Plan mode: letting the agent decompose hard tasks
update_plan makes the agent break a big goal into 5-10 actionable steps; you can interrupt and rewrite any step.
Read more →The full flow: letting AI co-write your tech blog
From topic → outline → research → first draft → revision → publish — where the agent fits at each step.
Read more →Generating weekly and daily reports with the agent
Let the agent scan this week’s commits, doc edits, and ticket status → produce a templated draft.
Read more →The complete agent toolset, organized
A cheat-sheet of six tool families with common usage — perfect for copy-pasting into agent system prompts.
Read more →Save any web page to your vault, in one prompt
`web_article_save` extracts the article, lands it as Markdown, files it in your vault.
Read more →Let the agent reorganize your messy vault
200 messy notes in Inbox? One prompt to classify, rename, and add frontmatter.
Read more →The Kition AI agent in one map: what it can really do
Not just another "chatbot that writes emails." It runs commands, patches files, browses, talks to MCP, edits tables.
Read more →Ready when you are.
Kition is a local-first AI workspace. Markdown documents, structured tables, and an AI agent — running on your own machine, against the model provider you choose.