Claude Basecamp: One npx Command Keeps Tests Green

Engineering leads running Claude Code across more than one repository know the pattern well. A test suite turns red overnight, a dependency bump sits in a pull request queue for weeks, and the issue tracker fills with duplicates nobody has time to close. Checking on all of it after the fact is not a system — it is a chore that gets worse with every repo added.

Kubernetes popularized one idea in infrastructure: declare the state you want, and a controller keeps reconciling reality to match it. "I wanted that for my codebase, so I built Claude Basecamp and I'm open-sourcing it today," the developer behind the tool wrote, describing it as a reconciliation loop for Claude Code — an application of that same declare-and-reconcile pattern to a repo's tests, dependencies, and issue backlog.

The babysitting breaks down into two separate failure patterns. One is reactive: a CI notification or a cron job tells you something broke, and you still have to go fix it yourself. The other is amnesia: context about a repo's goals, its past mistakes, and its half-finished sessions evaporates the moment a terminal window closes.

Install Basecamp with one command

The write-up describes installing Basecamp with a single npx command that starts a local, open-source dashboard for the repo.

Every check that follows runs against the repo's real signals — its actual test suite, its actual npm outdated output, its actual GitHub issues — not a simulation of any of them.

Four standing checks replace manual babysitting

"'Tests always green,' 'dependencies current,' 'the README documents every CLI flag' — say it once, Basecamp keeps it true, dispatches a fix run when it drifts, and only bugs you for the decisions that actually need a human," the write-up says, describing the four standing checks built into the tool.

Standing check	What it does
Tests always green	Runs the suite on a cadence; a failure dispatches a fix run that commits.
Dependencies current	Checks npm outdated; safe updates apply automatically, and major version bumps escalate to you.
Issue backlog triaged	Uses GitHub's gh tool to label issues and close stale ones.
Custom, in plain English	Type a goal such as "keep the tests green from now on" and Basecamp turns it into a standing check.

The write-up frames this as a trade: instead of a cron job that only pings you when something breaks, a standing check watches continuously and escalates just the decisions that need a human. "Checks run against deterministic local facts — your real test suite, real npm outdated — wherever possible, zero tokens spent checking," the author writes. "Drift launches a bounded, budgeted, approval-gated convergence run. Repeated failure escalates to a decision card on Home instead of retrying forever."

That is a meaningful difference for a team spread across several repos. A cron job reports status and waits for a human; a standing check is supposed to attempt the fix first and only interrupt you for the calls a script cannot make on its own.

Reflexes: an author-described immune system for pushback

The author calls one feature Reflexes and describes it as "an immune system for your AI." Basecamp mines every transcript for the moments you pushed back — an interruption, a flat "no, don't," a permission denial — and turns each into what the write-up calls an antibody.

Once that memory exists, the write-up claims every Claude Code session on the machine consults it before every Bash, Write, or Edit action — so a mistake corrected once is, in the author's words, "blocked machine-wide before it happens a third time."

Session Rescue: resuming work a dropped session left behind

A related feature, which the author calls Session Rescue, watches for a Claude Code session that dies mid-task. The write-up describes resuming the exact dead session — same session ID, full context — as a background run that finishes the job and commits.

Together, the two features target the amnesia half of the problem. The appeal for a team running several repos through Claude Code is less any single feature and more the idea of one place that remembers every open goal, every past mistake, and every dead session — instead of that context disappearing when a terminal window closes.

What the author claims about the safety model

None of this runs on someone else's server, and the write-up is explicit about the trust model. These are the author's own claims about the project, not an independent audit:

• Zero dependencies, about 4,000 lines of plain Node and vanilla JavaScript — "auditable in one sitting," per the write-up.
• Local only: the dashboard binds to 127.0.0.1, and the author says nothing leaves the machine and there is no telemetry.
• CSRF-guarded endpoints, with webhooks that use, in the author's description, "unguessable per-routine tokens."
• Sandboxed by Claude Code's own permission modes, and released public domain under the Unlicense, so anyone can fork it, audit it, or vendor it.

Worth weighing before wiring any of this into a real workflow: none of these security and safety claims come from an independent audit. A machine-wide reflex only helps if it generalizes the right lesson from one pushback, not an overcorrection from a single bad prompt.

The ambition behind it, and what's on the roadmap

The author admits the project outgrew its original goal: "this started as 'let me stop babysitting my tests and just make them stay green,' and it turned into something a lot bigger," the write-up says.

It adds an ambition for Basecamp to become, in the author's words, "the operating system for everything you do with Claude Code" — a destination the piece frames as where the project is headed, not something already finished.

The public roadmap stays short: a cross-repo resource governor for token budgets, per-routine cost guardrails, and more routine templates, built out as people run Basecamp against their own repo setups and report what breaks, as the author describes in the original write-up.

The cheapest way to test the idea costs one terminal line: point Basecamp at a repo where tests already flake and see whether the first standing check catches a regression your CI missed.