Persistent Context
Keep useful facts, decisions, and source material available for the next run.
coachix.
Agent practice
AI agent skillsfor repeatable workand recovery.
AI agent skills turn useful work into procedures an assistant can repeat: memory handling, deployment evidence, multi-agent handoff, failure recovery, and machine-readable artifacts.
Private knowledge is a control layer.
AI agent skills are reusable procedures that turn a successful run into something an assistant can repeat. A skill should carry the task shape, required inputs, allowed tools, stopping rules, checks, examples, and recovery notes that make the next run faster and safer.
For a private assistant, skills are the bridge between knowledge and action. Knowledge stores the source material and decisions. Skills say how to use that material in a repeatable workflow.
This page connects skills to private knowledge, MCP Tool Access, the console, and model routing so builders can package behavior instead of restarting from a prompt every time.
Console workflow
Apply repeatable procedures inside a visible assistant workspace.
MCP Tool Access
Connect files, browser state, APIs, and services through explicit tool boundaries.
What AI agent skills are
AI agent skills are reusable procedures that turn a successful run into something an assistant can repeat. A skill should not be a vague instruction like be careful. It should carry the task shape, required inputs, allowed tools, stopping rules, checks, examples, and recovery notes that make the next run faster and safer.
For a private assistant, skills are the bridge between knowledge and action. Knowledge stores the source material and decisions. Skills say how to use that material in a repeatable workflow: research, compare, write, test, deploy, verify, summarize, hand off, or recover from failure.
How to design a useful skill
Start with one workflow a person already repeats. Name the expected outcome, the sources the assistant may use, the tools it may call, and the point where the user must approve the next step. Then add a checklist that can be verified from output, not from the model claiming it was careful.
Good skills are narrow enough to test and broad enough to compound. A deployment skill can require build logs, production fetches, screenshots, sitemap checks, and a completion ledger. A research skill can require source freshness, primary-source preference, conflicting evidence notes, and a citation style.
Why skills help product understanding
The skills layer gives Coachix a concrete answer to a common agent-builder problem: how do you make an AI assistant reliable across repeated work instead of restarting from a chat prompt every time? The answer is to package behavior into procedures with evidence, memory, and boundaries.
Philosophy defines judgment, knowledge defines context, MCP defines tool boundaries, skills define repeatable work, and the console lets the agent run the pattern. This makes the site architecture legible to both search engines and builders.
What is an AI agent skill?
An AI agent skill is a reusable procedure that defines inputs, tools, checks, examples, and stopping rules for a repeated workflow.
How are skills different from prompts?
A prompt shapes one answer. A skill preserves a workflow so future runs can use the same evidence rules, tool boundaries, and recovery steps.
Where do skills fit in Coachix?
Skills sit beside private knowledge and MCP Tool Access: knowledge supplies trusted context, MCP defines tool boundaries, and skills define how the assistant repeats the work.
Work with this hub
Pair this page with the console, source context, tools, skills, and model routes when a private assistant needs to move from reading to action.
Open Console to apply the workflow / Private knowledge vault / MCP Tool Access / Latest LLM routes for model choice / Thinking patterns for agent builders