GenerateSaaS

Agent context

What a generated project ships for AI coding agents - AGENTS.md, the docs silo, and the bundled update and translate skills installed into every tool root - and how update refreshes and eject removes them.

Every generated project ships a complete working context for AI coding agents, whichever assistant you use. Three layers, all installed by init:

LayerWhereWhat it gives an agent
AGENTS.mdProject rootStanding instructions tailored to your stack: where things live, house rules, the gotchas agents trip on.
Docs silodocs/Feature and architecture reference citing real config keys and file paths - the facts an agent reads before it searches.
Skills<root>/generatesaas-update/ and <root>/generatesaas-translate/ in every skill rootupdate applies a staged boilerplate update as a guided 3-way merge - you decide every change. translate keeps every locale in step with your agent as the translator, no API key needed - see i18n.

Process skills are deliberately yours to write: the bundle carries only workflows that drive the project's own scripts, so it never competes with the conventions you give your agent.

Where the skills install

The same bundle is written to all five tool roots, unconditionally - which assistant you use is not a generation option:

.claude/skills/, .cursor/skills/, .agents/skills/, .gemini/skills/, .windsurf/skills/

Delete the roots you do not use, or leave them - they are inert text. Your own skills live safely beside the bundled ones: nothing in the lifecycle below ever touches a directory it did not ship.

Lifecycle

CommandEffect on the bundle
initInstalls AGENTS.md, the docs silo, and the skills into every root.
updateRefreshes the bundled skills in every root, always - even when no new version is staged. Writes only the bundled skill names; your own skills are never modified or deleted.
ejectRemoves exactly the shipped skill directories from every root, and nothing else.

On this page