Export

faf export writes the files your AI actually reads — AGENTS.md, .cursorrules, GEMINI.md, and more — straight from project.faf. One source, every surface, always in sync.

Write one

faf export --agents     # AGENTS.md

Every target

Flag	Writes
--agents	AGENTS.md
--cursor	.cursorrules
--gemini	GEMINI.md
--copilot	.github/copilot-instructions.md
--grok	wires grok-faf-mcp into .grok/config.toml
--llms	llms.txt (llmstxt.org shape) — the authored 6 Ws only
--conductor	Conductor config
--html	project.html — a visual render of project.faf
--card	MCP Server Card (.well-known/mcp/server-card)
--all	every format above

CLAUDE.md isn't here — it's bidirectional, not one-way. See Sync (soon).

Non-destructive, by design

Every writer wraps its output in a managed block:

<!-- faf:start -->
… facts from project.faf …
<!-- faf:end -->

• File doesn't exist → created, containing just the block.
• Block already there → only what's between the markers is replaced.
• File exists, no markers, hand-written → the block is prefixed; your content is preserved below it, byte for byte.
• Re-running with no changes → the file is untouched — idempotent.

faf owns the block. You own everything outside it. Enhance, never replace.

Under the hood

Every faf-authored file opens with its own fingerprint, so a re-run can tell its own prior output apart from something you wrote by hand:

<!-- faf: my-project | TypeScript | backend |  -->

That fingerprint is also the safety check: a markerless file that starts with it is faf's own legacy output and gets upgraded in place; a markerless file that doesn't is yours, and only ever gets prefixed.

---

Next: Custom rules — pin your own instructions so exports carry them forward