AI agents

Because stratt gives every repo the same verbs, an AI coding agent only has to learn one interface instead of rediscovering each project’s build and test commands. The stratt agents namespace makes that interface discoverable.

agents context

One command prints everything an agent needs to drive stratt here: a short orientation on what stratt is and how to use it, followed by the live command map for the current repo — detected stacks and the backend each verb resolves to.

stratt agents context

Point an agent at this once and it has the resolved commands, the relevant flags (--ci, --yes, --no-push), and the exit-code convention. Like doctor, it surfaces config or detection problems inline rather than failing, so the agent still gets whatever is knowable.

If you’ve configured a workspace root, agents context ends with a short workspace section: it notes that the repo lives in a stratt-managed workspace, gives the layout, and points the agent at stratt workspace list to discover sibling repos. The section is omitted entirely when no [workspace] is configured — there’s nothing to describe and no tree to list.

agents init and sync

stratt agents init adds a short stratt block to the repo’s AGENTS.md, telling any agent that reads it that stratt is in use and to run stratt agents context to learn the commands.

stratt agents init    # add the stratt block to AGENTS.md
stratt agents sync    # rewrite the block if stratt's boilerplate has changed

The block is a pointer, not a copy of the command map. It carries no repo-specific detail, so it rarely changes — the live map is always fetched fresh by agents context. The block is delimited by HTML-comment markers, which are valid Markdown, omitted from rendered output, and stripped from CLAUDE.md before a model sees them. sync finds the block by those markers and replaces it in place, leaving the rest of the file untouched.