For agents
A coding agent can pick tehas up from one address: /llms.txt on
this host. It says what tehas is, how to get work built by this factory, how to
set tehas up for another repository, and where the live data is.
Hand it to your agent
Paste one of these, with this host's address in place of <this host>:
Read http://<this host>/llms.txt, then write an issue this factory can build for: <what you want>
Read http://<this host>/llms.txt, then set tehas up for this repository. Ask me before deciding anything about secrets, accounts or who may authorize work.
Away from this host, start from the README of the public repository instead.
What an agent can read here
| Address | What it is |
|---|---|
/llms.txt |
The entry point: a summary and links, as plain text |
/<page>.md |
Any manual page as Markdown, for example /new-project.md |
/health |
Factory health as JSON; 503 when unhealthy |
/live/status.json |
The queue snapshot: repository, issues by state, paused or not |
/live/running.json |
The current or last cycle |
/live/console-<issue>.txt |
That cycle's console, with tokens and passwords masked |
All of it is read-only. Work is submitted and authorized in GitHub, as Operating describes.
What an agent may not decide
- Authorizing work.
tehas:readyspends compute and only counts from an account on the allow list. An agent writes the issue; a person labels it. - Merging. Nothing merges on its own.
- Secrets and accounts when installing: the bot's token, the allow list and the model endpoint are the owner's to choose.