🤝 Top Collaborators
🤓 Latest Submissions
RippleTrack
The Problem When a developer asks an AI coding assistant to "add a field to the user profile," the AI usually does exactly that - and nothing more. It edits the model, maybe touches the obvious API route, and stops. What it misses is everything downstream that silently depends on that model: the analytics pipeline that reads the field without a null check, the mock server fixture that's now out of sync, the test suite that will fail in CI two days later. By then, the original developer has moved on, and someone else is left debugging a "mystery" regression with no idea it traces back to one small requirements change. AI coding tools are excellent at writing the change you asked for, but blind to the ripple effects of that change across a real codebase. Our Solution RippleTrack is a multi-agent system, built on IBM Bob 2.0, that reads a requirements document (like a PRD) and a codebase together, and maps out every place a proposed change will actually touch - before a single line of code is written. Three specialized subagents run in parallel using Bob's Plan Mode: - Contract Detective - parses the PRD to extract the exact contract of the change (field name, type, optional/required, default value). - Code Archaeologist - traces the codebase's AST to find every consumer of the model being changed. - Test Archaeologist - finds every mock, fixture, and test that assumes the old shape of that model. An orchestrator merges their findings into a Ripple Map (a visual graph of what's connected to what) and a Risk Score (weighted by factors like untyped consumers, missing defaults, and stale fixtures). Only after this analysis does Bob switch to Agent Mode to execute: it writes one clean, reviewable feature branch for the human-facing code, and separately logs a machine-readable entry into a Signal Registry (`ripple_signals`) that downstream systems - analytics guards, mock syncs, CI - can consume directly. A lightweight CI script then performs Signal Sync, mounting those fixes automatically so the whole system goes from red to green without a human manually chasing every consumer. Who It's For RippleTrack is built for developers and small engineering teams working in fast-moving codebases - exactly the situation where "just add this field" quietly breaks three other things. It's especially valuable for teams already leaning on AI coding agents, since it plugs directly into that workflow instead of replacing it. What Makes It Different Most AI dev tools optimize for writing code fast. RippleTrack optimizes for understanding consequences before writing anything - treating impact analysis as a first-class step, not an afterthought. By splitting output into "what a human reviews" (a clean PR) versus "what CI consumes" (the Signal Registry), it keeps changes simple for people while still closing every downstream gap automatically. The result: one clean commit in the git log, and a codebase that was never actually vulnerable to the regression in the first place.
27 Sep 2026
