> ## Documentation Index
> Fetch the complete documentation index at: https://docs.splox.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Reference

> The exact shape of a harness — programs, agents, hooks, tools, skills, memory, evals — written for your agent to read

<Info>
  **This section is for your agent.** Everything in it is precise on purpose:
  signatures, file layouts, the rules the platform enforces. You do not need to
  read it to use Splox. When you ask your agent for something and it needs a fact
  from here, hand it the page — every page has *Copy page* in the top-right, and
  the URL alone is enough for an agent with a browser.

  What to ask for is in [Ask](/ask/overview). How to see what changed, and go
  back, is in [Look inside](/inside/overview).
</Info>

A harness is a git repository checked out on the machine at `~/harness`. The
platform reads it at the moment it needs to know something — who an agent is,
what a hook does, which tools exist — and never keeps a parsed copy. That is why
a change is live the moment a file is saved, and why a push is a version.

What is in it:

| Path                                   | What it declares                                                         | Page                            |
| -------------------------------------- | ------------------------------------------------------------------------ | ------------------------------- |
| `programs/<name>/main.py`              | A program: its agents, and `handle(msg)` for the one that answers a chat | [Programs](/reference/programs) |
| `agent(...)` in that file              | Prompt, model, provider, tools, skills, limits                           | [Agents](/reference/agents)     |
| `programs/<name>/hooks/<point>.py`     | What the loop does at one point of a turn                                | [Hooks](/reference/hooks)       |
| `tools/<file>.py`                      | Functions offered to the model by name, schema from the signature        | [Tools](/reference/tools)       |
| `skills/<name>/SKILL.md`               | A procedure the agent reads when a task matches its description          | [Skills](/reference/skills)     |
| `~/memory/` on the machine, not in git | What the agent keeps between chats                                       | [Memory](/reference/memory)     |
| `evals/cases`, `evals/graders`         | How a version is checked against the last one                            | [Evals](/reference/evals)       |

Two things the reference assumes throughout: a program that calls other agents
does so with the shape in [Sub-agents](/reference/subagents), and the worked
examples for a router, a nightly job and a fan-out are in
[Patterns](/reference/patterns).

<CardGroup cols={2}>
  <Card title="Programs" icon="folder" href="/reference/programs">
    What the platform reads out of `main.py`, what it never opens, and how a program calls its agents.
  </Card>

  <Card title="Agents" icon="robot" href="/reference/agents">
    Every parameter of `agent(...)`, what a run inherits, and what a sub-agent does not.
  </Card>

  <Card title="Hooks" icon="anchor" href="/reference/hooks">
    The points of a turn, what each receives, what each may return.
  </Card>

  <Card title="Tools" icon="wrench" href="/reference/tools">
    A function becomes a tool: naming, schema, errors, what the model is shown.
  </Card>
</CardGroup>