.png&w=256&q=75)
🤝 Top Collaborators
.png&w=640&q=75)
🤓 Latest Submissions
Uplift: Know What a Change Will Break
Every developer has pushed a change that broke something unexpected. The bug was real, the impact was predictable, but no tool caught it in time. Uplift solves this. The problem: Code changes have hidden blast radii. A single function change can silently break downstream callers, route contracts, and edge cases across modules. Static analysis flags syntax; nothing maps the live impact of a change before it ships. What Uplift does: Uplift is a code change impact analysis engine with a visual dashboard, a CLI, and a native MCP server built specifically for IBM Bob. Given any code change, Uplift runs a four-stage pipeline: Predict, Prove, Repair, Verify. Predict: Uplift builds a call graph from the diff, identifies every affected symbol, and assigns a verdict (will break, might break, safe) to each candidate. Prove: For each predicted break, Uplift generates a proof test that passes on the base commit and fails on the changed head. No guessing; every prediction is backed by a reproducible test. Repair: Parallel IBM Bob worker agents, each sandboxed to a single module with strict file-level edit restrictions, run concurrently to fix every broken module. Users, orders, payments, and core are repaired simultaneously. Verify: The full test suite reruns. Zero regressions is the exit condition. Target users: Backend engineers and DevOps teams shipping Python services who want confidence before merge. The workflow starts in the terminal with uplift graph, integrates into CI via a GitHub Action that posts an impact comment on every pull request, and is fully accessible through the dashboard at a public URL. What makes it unique: Uplift is the only submission that exposes a native MCP server (uplift mcp) that IBM Bob modes call directly. Four Bob custom modes (impact-analyst, prover, migration-planner, verifier) each hold a dedicated MCP permission group. Three repair worker modes run in parallel with fileRegex restrictions that prevent cross-module edits. The pipeline is fully reproducible: • Total tests: 339 • Engine tests: 86 • Sample-app tests: 52 • Dashboard tests: 201 • Verifier checks: 104 • Verifier failures: 0 • Verifier warnings: 0 The verifier runs 104 checks covering schema validation, Bob evidence, scenario branches, proof counts, accuracy metrics, and deploy config. All 104 pass with zero failures and zero warnings. The pydantic migration scenario demonstrates the system at scale: Uplift scans a pydantic v1 codebase, builds a migration catalog from the official migration guide PDF, dispatches parallel Bob workers to repair each module, and produces a full migrate report with before/after test counts for every worker lane. Before repair: • Failures: 8 • Collection errors: 6 After repair: • Tests passed: 46 • Failures: 0 • Collection errors: 0 All visible in the dashboard Migrate view. Uplift is not a demo of what Bob could do. It is a working product, deployed, with real evidence, real proofs, and real repairs committed to the repository.
27 Sep 2026
