A PR walk that ends with a next move.
Arbor reads the changed code, builds the graph around it, traces the downstream paths, and posts a compact note that says what might break, what is unknown, and what to test first.
No guessing loop. The walk is structural, and unresolved edges stay visible.
From diff to breakage brief in four steps.
Arbor keeps the hard analysis deterministic, then turns the result into review context that humans and agents can both use.
Read the PR
Arbor starts from the diff, changed symbols, imports, and framework entry points. It builds the local map reviewers usually reconstruct by hand.
// PR #184
changed:
- billing/checkout.ts
- billing/plan-id.ts
entrypoints:
- POST /checkout
- POST /webhooks/stripeAgent idea: Arbor as the pre-edit context server.
Before an AI agent edits a repo, it can ask Arbor for the PR's breakage brief. Arbor returns changed paths, likely side effects, unknown edges, and the first test to write. That keeps agent work grounded in the code graph, not just the prompt.
Arbor watches the parts of a small product that usually surprise you.
The model is built around product breakage: billing weirdness, auth regressions, background jobs, webhook retries, release scripts, and data writes.
Likely breakage path found. Add a test or patch before merge.
Path touches something fragile. Human review should focus here.
No obvious break, but the change is adjacent to important behavior.
No meaningful downstream path found. Unknowns are still listed.
Agent workflow: ask Arbor for the brief, let the agent write the missing test or constrained patch, then rerun the walk before merge.
Watch Arbor turn a tiny diff into a useful review plan.
The point is not another giant report. The point is one PR comment that tells a human and an AI agent where the change reaches, what might break, and which test or patch should happen next.
For the solo dev
You get the side effects you did not have time to trace manually: billing, auth, jobs, emails, CLI commands, and shared helpers.
For AI agents
Agents can read the Arbor brief before they keep editing, then limit their next patch to the risky path and generate the missing test first.
For tiny teams
Reviewers see the exact regression to test instead of re-learning the whole codebase in the PR thread.
14 languages. Real AST parsing.
Syntax-accurate parsing for every supported language. The graph is built from actual function definitions, call sites, and imports.