The AI use case

Your .md files are
the best prompts
you'll ever write

AI coding agents already read your repo. Cursor, Claude Code, Copilot. They all index your files for context. The question isn't whether AI will read your repo. It's whether it'll find specs, or nothing.

Same AI. Different repo. Wildly different results.

The only difference is whether your repo has .md files with real specs.

Without .md specs in repo
AI reads raw code, guesses intent
Generated code doesn't match your patterns
AI invents API routes that don't exist
Every prompt needs 3 paragraphs of context
Specs live in Notion. AI can't see them.
With .md specs in repo
AI reads your specs, knows the plan
Generated code follows your architecture
AI references real endpoints from your docs
"Build what's in docs/auth-prd.md" just works
Specs live in the repo. AI has full context.

Write the docs AI actually reads

CLAUDE.md, AGENTS.md, architecture docs, PRDs: write them collaboratively in verso.md, and every AI tool that touches your repo gets the full picture.

CLAUDE.md
AI Context
1# CLAUDE.md
2
3## Project overview
4Collaborative markdown editor built with
5Next.js, Supabase, and Yjs for real-time.
6
7## Architecture
8- App Router with Server Components
9- Supabase for auth + Postgres
10- PartyKit for Yjs WebSocket syncSarah
11
12## Conventions
13- Use `src/lib/` for shared utilities
14- API routes in `src/app/api/`- Components in `src/components/ui/`

The spec → code → spec loop

Docs and code evolve together. AI keeps both in sync.

Write the spec
In the repo, as .md
AI reads it
Full context, zero prompting
AI generates code
Matches your spec exactly
Update the spec
In the same PR as the code

“Build what's in the spec”

When your specs are .md files in the repo, talking to AI becomes trivially simple. No copy-pasting. No context windows. Just point at the file.

Terminal
$ claude "Implement the auth flow described in docs/auth-prd.md"
Reading docs/auth-prd.md...
Found spec: OAuth via /api/auth/[provider], session store in Redis, refreshToken rotation with PKCE.
Reading CLAUDE.md for conventions...
Creating:
+ src/app/api/auth/[provider]/route.ts + src/lib/auth/oauth.ts + src/lib/auth/refresh-token.ts + src/lib/auth/pkce.ts ~ src/lib/auth/session.ts
All files follow conventions from CLAUDE.md — utilities in src/lib/, API routes in src/app/api/.
$ claude "Update the PRD to mark OAuth as implemented"
Updated docs/auth-prd.md — marked OAuth section as ✓ Done, added implementation notes with actual file paths.

Every .md file is an AI multiplier

The more context you put in your repo, the smarter every AI tool becomes.

CLAUDE.md / AGENTS.md

Write agent instructions with real-time collab. They're immediately available to every AI tool that reads your repo.

Specs as prompts

"Build the auth flow described in docs/auth-prd.md". A well-written spec IS the prompt. No copy-pasting into chat windows.

Onboard AI like an engineer

Same getting-started guide, same architecture docs, same conventions file. AI reads them all, just like a new hire would.

Guardrails in markdown

Document what NOT to do. AI agents read your ADRs and know why you chose Redis over Kafka. They won't suggest Kafka.

Close the loop

AI reads the spec → generates code → you update the spec in the same PR. Docs and code never drift apart.

Review the whole picture

PRs contain the spec change AND the code change. Reviewers see the full story, not just a diff.

The best thing you can do for your AI tools is write great docs and keep them in the repo.

verso.md is the editor that makes this effortless.

Ready to try it?

Sign in with GitHub and start writing docs that live next to your code.

Get started, it's free