Trace Forge AI

TraceForge AI — ContractGuard is a production-oriented developer workflow application that orchestrates the complete lifecycle of API contract investigation and repair. Its tagline — From Bug Report to Verified Fix — describes the system precisely: it takes a bug report as input and produces an independently verified, evidence-backed, human-approved fix as output, with every decision recorded and traceable. The Problem Full-stack applications suffer from API contract drift. Over time, the frontend implementation, backend handler logic, automated test suite, and API documentation become subtly inconsistent with each other. When a failure surfaces, developers must manually cross-reference multiple layers of the codebase, reproduce the failure, reconstruct the original intent, coordinate changes across components, and confirm that the correction resolves the complete problem. This investigation process is fragmented, error-prone, and nearly impossible to audit after the fact. The Solution ContractGuard replaces this fragmented workflow with a structured, tool-assisted pipeline consisting of ten coordinated stages: Bug intake — Developers submit a bug report with expected behavior, observed behavior, and reproduction steps through the React dashboard. Workspace isolation — The Rust backend creates a dedicated, immutable working copy of the sample project, protecting the original fixture from modification. Baseline reproduction — The verification engine executes the existing test suite against the broken fixture, capturing failing test identifiers, exit codes, and logs as objective evidence of the defect. IBM Bob investigation — A structured prompt directs Bob to read the API documentation, OpenAPI specification, frontend source, backend handler, and existing tests. Bob uses parallel subagents for independent frontend, backend, and contract investigations, then synthesizes findings into a validated diagnosis.json artifact, an investigation summary, and a fix plan. Artifact import and validation — The backend validates every imported artifact against a published JSON schema, checks source-file references, enforces workspace path boundaries, and rejects malformed data without altering the investigation's current valid state. Human approval — The developer reviews the findings, proposed corrections, affected files, regression-test plan, and associated risks. Approval is recorded with a content hash of the plan; any material modification requires renewed approval. Bob-assisted implementation — A second structured prompt instructs Bob to apply only the approved corrections in the isolated workspace, add regression tests, and produce an implementation summary and code diff. Independent verification — The backend's verification engine re-executes the full test suite using Tokio process management against a predefined allowlist of commands, enforcing execution timeouts, output-size limits, and path canonicalization. Verification status is determined entirely by actual test exit codes. Contract Evidence Graph — Each investigation builds a persistent, queryable graph connecting requirements, frontend findings, backend findings, root causes, approved fixes, and passing tests. Missing evidence links are displayed explicitly rather than fabricated. Final report generation — A complete investigation report is produced in HTML and JSON, covering all nineteen required sections including the original bug report, root causes, approved plan, code diff, regression results, evidence graph, and productivity measurements.