lfm: landing pages

lfm turns one page template, its config and assets into a static output/ folder, and refuses to ship pages that are broken, slow, inaccessible or full of generated defaults. Its own skills carry the process; this one gets you there.

1. Have the binary

Run lfm version. If the command is missing, install it and run lfm version again:

curl -fsSL https://lfm.cash/install.sh | sh

The same script is at https://raw.githubusercontent.com/Anonyfox/lfm/main/install.sh. With Go 1.26 or newer instead: go install github.com/Anonyfox/lfm/cmd/lfm@latest. The installer puts the binary in ~/.local/bin; if that is not on PATH, call it as ~/.local/bin/lfm. When neither works, tell the user what failed and stop: a page made without lfm is not this skill.

2. Have a project

• A folder with lfm.json is a project: work there.
• Otherwise create one in a new folder named after the offer, lfm init <offer-name>, then cd into it. Inside an unrelated repository, ask where the page should live; without anyone to ask, use a new subfolder of the current folder. Never initialize into a folder holding other work.
• Set language in lfm.json to the language the buyers speak (de for buyers in Germany).

3. Install the process skills

Run lfm skills install in the project. It installs lfm-research, lfm-design, lfm-implementation, lfm-book, lfm-fiction, lfm-printable, lfm-spreadsheet, lfm-document and lfm-deck where your agent loads skills and says where. If they do not appear in this session, read each one when its phase starts with lfm skills research, lfm skills design, lfm skills implementation, lfm skills book, lfm skills fiction, lfm skills printable, lfm skills spreadsheet, lfm skills document and lfm skills deck: they match the installed binary exactly. The last six are for what the page sells, in deliverables/book/, deliverables/printable/, deliverables/spreadsheet/, deliverables/document/ and deliverables/deck/.

Compile never writes skill files. After upgrading lfm, run the printed lfm skills install command for each folder with older copies. Edited copies require --force to replace.

4. Follow the compiler

Run lfm compile. The report ends with Next (<phase>): … and the skill for that phase. Do that step, compile again, and repeat until it says deploy:

1. research: real buyer voices and competitor pages, verbatim, into research/.
2. design: concept.md, every decision cited to the research.
3. implementation: the page from concept.md.
4. fix and review: 0 errors, 0 warnings, then lfm compile --release and every saved frame.
5. deploy: upload output/ as is. lfm does not host pages; ask the user where it goes.

lfm compile --json gives the same result as one JSON object, with next.phase and next.skill.

lfm check <url> --json checks any existing page, from any framework, with the same page rules. Add --project <dir> for project-only rules; omit the target to check output/ and detect stale files. Release checks require Chrome, Lighthouse and loopback sockets; run outside a sandbox that blocks them. Timing warnings use up to three measurements; informational diagnostics and third-party rows do not block release.

Rules

• Do not skip research or design to save time: the page without them is the generic page lfm warns about.
• Ask the user only what they decide: product, price, delivery, the page's language, legal pages and the payment link. Start research while waiting; the design skill records what is still open.
• Edit the template, lfm.json and assets/; never edit output/.
• Each finding says how to fix it; lfm rules <id> explains one.