powerbash.github.io

The site at powerbash.org — landing page and the full documentation.

Working on it

There is no build step and no dependencies. Open index.html in a browser:

xdg-open index.html      # or: open index.html
make serve               # or over http, if you need absolute paths to resolve

Everything is hand-written HTML and one stylesheet. The page makes zero external requests — no fonts, no CDN, no analytics — and runs no code: the only <script> on either page is an application/ld+json block, which is structured data for search engines and is never executed. The prompt configurator on the landing page is checkboxes and :checked sibling selectors, not JavaScript. Keep it that way.

make check enforces the parts of that which are mechanical, and CI runs the same target:

make check      # sitemap covers every page, skill digests match, no remote assets
make og         # re-render og.png from og.svg (needs chromium)

.nojekyll tells GitHub Pages to serve the tree verbatim rather than running it through Jekyll. It is also what makes .well-known/ reachable.

Colors

The palette is not invented. Every segment color in css/main.css is the color powerbash actually emits, read off a real prompt render — see the comment at the top of that file for the tput codes each one corresponds to.

Content

The documentation lives here, not in the app repo. Adding a user-visible setting to napalm255/powerbash means updating three files in this repo:

File	Why
docs/index.html	the documentation itself
.well-known/agent-skills/powerbash/SKILL.md	what agents install and configure powerbash from
.well-known/agent-skills/index.json	the digest there is a SHA-256 of SKILL.md

make check-skills fails when the digest and the file disagree, so the third one is hard to forget. Recompute it with:

sha256sum .well-known/agent-skills/powerbash/SKILL.md

New pages additionally need an entry in sitemap.xml — make check-sitemap fails otherwise — and usually a line in llms.txt.

Agents and search engines

The files that exist only to be read by machines, and what each is for:

File	Purpose
robots.txt	crawl rules, AI-crawler rules, Content-Signal, and the sitemap pointer
sitemap.xml	the canonical URL of every page
llms.txt	an annotated map of the site for language models (spec)
.well-known/agent-skills/	an installable skill, per the Agent Skills Discovery RFC
og.svg / og.png	the social preview card; the SVG is the source, the PNG is what ships

The content signal is search=yes, ai-input=yes, ai-train=yes — everything here is MIT-licensed documentation for a free script, and the point is for agents to be able to recommend and install it.

Some checks at isitagentready.com will never pass, and should not: MCP server cards, OAuth discovery, and the commerce protocols all describe an API and a checkout that a static docs site for a shell script does not have. Publishing them would advertise services that do not exist.

License

MIT — see LICENSE.