Robert Boulos, Snappy
Claude Code work, in your repo
I work in Claude Code all day.Bring me your repo.
I've been using Claude Code since it came out, and I help developers and founders who aren't AI-native get real value out of it. We open your codebase live on your screen, pick one outcome for the week, and ship it together.
target.md
Fix checkout recovery so failed payments return to the right plan.
you keep
- Repo map
- kept in CLAUDE.md
- Skill file
- checked into /skills
- Verification
- typecheck + browser
- Handoff note
- next moves written down
01
fit call
pick the stuck outcome
02
live week
3 private sessions
03
keep it
repo-owned operating notes
where the week starts
The whole repo isn't the problem.
It is usually one seam: the route nobody wants to touch, the test no one has written, the agent prompt that works once and then drifts. The week is about finding that place and making it legible.
Built fast with AI and now sitting on rubble? Start with a repo audit and find out what you actually have.
A feature branch has been open too long
I shrink the target until it can ship
Claude Code keeps losing the thread
I write the repo rules into durable skills
Nobody trusts the backend changes
I add the check that proves the route still works
AI built half of it and you're not sure what you've got
I audit the repo and name what's load-bearing and what's slop
Tokens burn faster than work ships
I set up the loops that get work done without going bankrupt on tokens
Three things you keep when I leave
Agent run log
Claude Code agents wired to your repo
During the build week, agents work inside your repo: loading CLAUDE.md, running tests, fixing stuck features, screenshotting the dashboard, writing handoff notes. Every run is logged. When the week ends, the agents stay clipped to your repo.
Skill file
Plain-English playbooks you keep in the repo
Every agent follows a skill file checked into your repo: which routes to call, which tools to use, what to do when a test fails, what it has already learned from past runs. Open any file, read exactly what the agent does, change a rule yourself.
Owned infrastructure
Runs, routes, and checks land in your stack
Your GitHub, your database, your hosting. State from every run, every deployed route, every verification check is written to accounts you own. When the week ends, you open your own dashboard and the whole build is sitting there.
Project Agent
Running tasks on your behalf
What can I help you build?
Try one of these
Your Backend
5 tables · 10 routes · 627 runs · 412 checks
| id | agent | target | tools | outcome | time |
|---|---|---|---|---|---|
| 627 | Code Fixer | app/api/contacts/route.ts | 4 | patched | 2m ago |
| 626 | Browser QA | /dashboard | 4 | pass | 14m ago |
| 625 | Test Runner | vitest + playwright | 4 | pass | 28m ago |
| 624 | Repo Context | CLAUDE.md + tree | 3 | loaded | 1h ago |
| 623 | Handoff Writer | docs/handoff/latest.md | 3 | written | 2h ago |
Agent run log
Claude Code agents wired to your repo
During the build week, agents work inside your repo: loading CLAUDE.md, running tests, fixing stuck features, screenshotting the dashboard, writing handoff notes. Every run is logged. When the week ends, the agents stay clipped to your repo.
Project Agent
Running tasks on your behalf
What can I help you build?
Try one of these
Skill file
Plain-English playbooks you keep in the repo
Every agent follows a skill file checked into your repo: which routes to call, which tools to use, what to do when a test fails, what it has already learned from past runs. Open any file, read exactly what the agent does, change a rule yourself.
Owned infrastructure
Runs, routes, and checks land in your stack
Your GitHub, your database, your hosting. State from every run, every deployed route, every verification check is written to accounts you own. When the week ends, you open your own dashboard and the whole build is sitting there.
Your Backend
5 tables · 10 routes · 627 runs · 412 checks
| id | agent | target | tools | outcome | time |
|---|---|---|---|---|---|
| 627 | Code Fixer | app/api/contacts/route.ts | 4 | patched | 2m ago |
| 626 | Browser QA | /dashboard | 4 | pass | 14m ago |
| 625 | Test Runner | vitest + playwright | 4 | pass | 28m ago |
| 624 | Repo Context | CLAUDE.md + tree | 3 | loaded | 1h ago |
| 623 | Handoff Writer | docs/handoff/latest.md | 3 | written | 2h ago |
What clients have said
From founders who've been through the process.
The honest alternative is doing it alone.
The week with me
Grinding it alone
Time to production
1–3 weeks
Working software on your accounts.
Open-ended
The branch that has been open for weeks.
Your investment
Hundreds
A cohort seat, or the private week at $900.
Burned tokens
Weeks of trial and error, nothing shipped.
What you keep
Everything
Code, agents, skill files, documentation.
A maze
Half-built agent code you are afraid to touch.
How it goes
Live, together
We fix it on your screen, in your repo.
Solo
You, the docs, and guesswork.
Between sessions
Agents keep working
Tests run and context stays warm for the next session.
You lose the thread
Every session starts from scratch.
The first session is free, on your actual codebase.
the accelerator
Run Claude Code the way I do. Live, in a small group.
I've been in Claude Code since it came out, working real client repos with it every day. The Accelerator is that, as a live cohort: a handful of builders, one focused week, each of us in our own codebase. You buy a seat, you do the work, you keep the setup and walk away. No retainer, no dependency.
Day one
each of us picks the one outcome we want by Friday and wires Claude Code to our own repo
Midweek
turn the manual move you keep redoing into a skill or an agent loop, live
Day five
ship the outcome and write the operating notes you keep
Cohorts are small and I run them rarely. Prefer it 1:1? The private week is still on the table.
A free session on your project.
We'll open your codebase together and start working on it. If it's a fit, the usual next step is the Claude Code Accelerator: a small live cohort, one focused week.
what to expect
Architecture mapped, bottlenecks identified
Live building on your codebase, screen to screen
A clear plan for what to build next
Email me
robert@snappy.aiBook a call
calendly.com/robertboulosBefore you book
What founders ask before the first call.
Free, no commitment, on your screen.