docouture treats a documentation site as something a coding agent both reads and writes, not just something a human authors and an agent happens to stumble into. Three things make that real: the site publishes itself in a form an agent can read directly, a set of agent skills can write and maintain it, and a ledger keeps that writing honest instead of letting it decay into "AI-authored once and never touched again."
Reading the site — llms.txt
Every build publishes llms.txt (a Markdown index) and llms-full.txt (a full
aggregated Markdown dump) at the site root, the llms.txt
convention — so an agent can load the whole site’s content directly instead of
scraping rendered HTML. Configuration (the summary blockquote, excluding specific
pages) lives with the rest of this site’s build-time integrations, not here — see
Integrations.
Writing the site — agent skills
npx skills@latest add InditexTech/docouture --all installs a set of agent skills that
plan a site’s structure, draft its first content from whatever the repo already has,
and keep it in sync as the repo changes. Skill installation is deliberately separate
from, and never scaffolded by, docouture new/docouture upgrade themselves — a
repository picks up, updates, or drops these skills on its own schedule, independent of
scaffolding.
| Skill | What it’s for |
|---|---|
|
The entry point for a repo with no site yet: scaffolding it if needed, planning its module/page structure, and drafting first content from whatever the repo actually has. This site was drafted this way — see About. |
|
The mechanics once a page/module needs to exist: the playbook, the component descriptor, the four names that must agree, the home-page vs. content-page patterns. |
|
What to actually write on each page: purpose, section skeleton, per-section instructions, a copyable AsciiDoc starting point and a quality checklist — for the six standard sections (Overview, Getting started, Guides, Reference, Additional information, Contributing) and the home page. |
|
AsciiDoc itself: the language, |
|
The re-entry point once a site exists and is evolving: figuring out what a diff/PR changed, whether it needs a docs update, and keeping `AGENTS.md’s documentation-state ledger honest. |
|
Cutting a release on a |
Setting up the skills
The skills are installed with skills, a
third-party CLI for the open agent-skills ecosystem (not part of docouture itself) that
works across dozens of coding-agent harnesses — OpenCode and Claude Code among them.
-
Install them from the repository root:
npx skills@latest add InditexTech/docouture --all--allinstalls every skill listed above to every agent harness the CLI finds; pass--skill <name>one or more times to install only specific skills, or-a/--agent <agent>to target one harness instead of all detected ones. -
Let it detect your harness. The CLI scans the repository for installed coding agents (OpenCode, Claude Code, Codex, Cursor, and others) and installs to each one it finds; if it finds none, it prompts for which to install to instead.
-
Know where the files land. Each harness reads skills from its own project-scoped directory — for example
.claude/skills/for Claude Code or.agents/skills/for OpenCode, Codex, Cursor, and several others. By default the CLI symlinks every harness’s copy back to one canonical install rather than duplicating the files, so a later update only has to touch one place. -
Verify the install:
npx skills list -
Reload the agent session. A skill’s
descriptionis what lets an agent reach for it on its own; that only takes effect once the agent (re)reads its skill directory, so restart or reload the session after installing before expecting a skill to fire automatically. -
Keep them current as this repository’s skills evolve:
npx skills update
See the skills CLI’s own documentation
for the full command reference (removing skills, global vs. project scope, and the
complete list of supported agent harnesses).
Staying honest — the documentation-state ledger
docouture-documenting-changes maintains a ## Documentation state table in
AGENTS.md: which pages were drafted straight from the repo’s own code (and from
what), versus hand-written — so a later pass can tell "still matches the code" apart
from "needs a look" without re-deriving every page from scratch. This site’s own
AGENTS.md carries exactly that table.
docouture doctor advisory-checks that AGENTS.md is still present (not the skills
themselves — see CLI, installing and maintaining them is
entirely the skills CLI’s job), so a repository that loses its ledger notices rather
than silently drifting into undocumented territory.
This site you’re reading was itself scaffolded and drafted this way — see About.