# For agents An agent that drives rietx reads the rietx skill, a protocol and a set of reference files in the agentskills.io format. The protocol says what to free in what order, what to check before believing a number, and which numbers may be quoted. An agent reads it whole before its first fit. The reference files are lookups it loads when a task needs one, such as the API index, the diagnostic codes or a series of patterns. [Calling rietx from a program](programs.html.md) gives the Python surface the skill drives: the exceptions, the answer types, an answer’s JSON form and `capabilities()`. ## Installing the skill The package carries the skill. This command installs it into the current project: ```console $ rietx skill --install . ``` The copy goes in `.agents/skills/rietx/`, which several harnesses read, and a link goes in each requested harness’s own directory. The command prints both paths, then two lines to add to the project’s `AGENTS.md` or `CLAUDE.md`. It does not write those lines itself. `--agent NAME` adds a harness, and `rietx skill --list-agents` lists the ones it knows. `--user` installs for your user account instead of one project. ## Reading the skill without installing it `rietx skill --print` prints the protocol, and `rietx skill --print all` prints every file. The same files are published at [rietx.org/skill/rietx/SKILL.md](https://rietx.org/skill/rietx/SKILL.md), one file at a time, for a harness that has network access and no install.