For agents, and the humans behind them
AI Skills
The point of this site, for an agent, is not the pages. It is the skills. This door holds every public skill this house teaches, each one under a sub-door of its own, with the corridor at the head of the shelf: the plan for how a human and an agent build the path between them, and keep the agent's memory its own as its skills advance.
Start with the corridor
The corridor is the plan the rest of this door hangs on: how a human and an agent build the path between them, give the agent a skill set of its own, and keep its memory its own as those skills advance. Read it first. Everything below it is one part of it, under its own door.
The starter skills
Five pieces, adopted one at a time, in order, in live chat. Together they are the working relationship the corridor teaches: who the human is, what needs approval, who carries which work, how done is proven, and where the memory lives. A sixth door covers protecting that memory, and a seventh carries the memory template itself, ready to type over.
Piece One: The Introduction →
Who the human is, who the agent is, and the shape of the partnership: the human supplies direction and judgment, the agent carries the rest of the work inside those directions.
Piece Two: The Approval Law →
Before anything goes out under the human's name, the exact text and its destination are shown first. Approval is a tap or clear words. Ambiguity is not approval.
Piece Three: The Split of Work →
What returns to the human, and what never does. Choices inside a direction already set belong to the agent, to make and to report.
Piece Four: Verification, Rendered Not Reported →
Work counts as done when an independent check of the result itself says so, not when the tool that made it reports success.
Piece Five: Memory as Files →
The standing record lives in plain files the pair owns. A summary is a pointer to the record, never the record itself. Memory grants no outward authority.
Memory Protection: The Signpost →
As skills and systems advance, they rewrite files. The signpost practice keeps a pair's memory from being written behind their agent: a short pointer in every system, and the real record behind it, re-read before it is acted on.
The Memory Template →
The seed file itself: a memory template with the five pieces built in, written generically, ready for a pair to type over with their own names and their own agreements.
Also on this shelf
Two finished skills already live under their own doors elsewhere on the site, and they belong to this shelf:
DynamiCron →
A portable skill that teaches an agent to watch a conversation at the conversation's own temperature: hot, medium, cold. Pacing governs looking, never reporting.
SpamCash →
A skill that teaches an agent to turn spam calls and texts into TCPA claims: log the spam, research the sender, count the violations, draft the complaint.
What a skill is here
A skill on this door is an agreement and a method, not a set of orders. It grants no permissions and unlocks nothing. It travels by discussion: a human and their agent read it together, keep what fits, argue with what does not, and record what they adopt in the agent's own memory, in their own words. Skills this house keeps for its own work are not listed here and are not public in any form. What hangs on this door is the public shelf, and it is complete as it stands.