# Working with agents

> ATH is the record and the evidence layer. The model is yours to choose.

Website: https://ath.fit/docs/agents/
Markdown: https://ath.fit/docs/agents.md

ATH v0.4.3 · MIT license. Documentation source: [cd89f72](https://github.com/mwhitham/ath/blob/cd89f72eceae3961ef1a05694226515c9e4321d2/skill/SKILL.md). If your installed version differs, check its command help.

Checked 27 September 2026: v0.4.3 is available on npm.

## Give your agent the instructions

ath init installs the skill in .agents/skills and any existing agent folders. Open your agent in that same folder. Use ath skill to check the installed copies, or ath skill install to install or refresh them without upgrading the CLI.

Starting in v0.4.2, ath --update also refreshes an existing skill in the current folder after checking or updating ATH. Run it from each ath folder whose skill you want to refresh.

```text
ath skill
ath skill install
```

- [Update ATH and its skill](https://ath.fit/docs/update/)

## Read before reasoning

Start with ath stats. Use structured output, preserve source attribution and name missing data. A day without a recorded session is not necessarily a rest day.

```text
ath stats --json
ath predict fran --json
```

## Keep the future out of the evidence

--as-of hides observations after a chosen date. This lets an agent reason from a historical view. Use ath backtest to replay historical results with gateway models and write a separate evaluation report.

```text
ath predict fran --as-of 2026-06-01 --json
```