FabricPilot

Problem: enterprises running Hyperledger Fabric accumulate real technical debt — chaincode written against the deprecated raw shim API, client applications built on the legacy HFClient SDK. Migrating this code is risky: a broken endorsement policy or an incompatible chaincode invocation can take down a live, permissioned network that multiple organizations depend on. General-purpose AI modernization platforms exist (IBM watsonx Code Assistant, AWS Transform, Modelcode's Morph, vFunction), but none of them are specific to Hyperledger Fabric — they treat code as generic text, with no understanding of endorsement policies, channel membership, or chaincode lifecycle. The closest Fabric-specific prior work we found is either infrastructure migration guidance (e.g., IBM's own IBP-to-Fabric-support migration documentation) or manually-written community tutorials for one-off chaincode ports — not an automated, risk-aware tool. No existing solution combines legacy-pattern detection, sandboxed validation, and an immutable on-chain audit record for a Fabric modernization change. Compounding this, existing commit-provenance tools sign changes with an external key stored off-chain — meaning a blockchain's own guarantee of immutability is never actually used to protect the record of the change itself. Solution: FabricPilot turns IBM Bob into a Hyperledger Fabric modernization specialist through a custom Bob mode and a purpose-built MCP server exposing four tools. `analyze_repo` statically scans a Java + Fabric codebase for legacy SDK patterns and produces a risk-scored impact report. `run_sandbox` mirrors the target network in an isolated Docker Compose environment and applies the proposed change there, running real smoke tests, before anything touches production. Only if the sandbox passes, `write_ledger_stamp` submits a real Fabric transaction to a dedicated audit chaincode, permanently recording the change's hash, risk level, and test results on the network's own ledger — not as an external signature, but as a transaction endorsed, ordered, and committed exactly like any other business transaction. `query_ledger_stamp` lets anyone verify that record later, live, against the blockchain itself. Target users: platform engineers and blockchain architects at enterprises running Hyperledger Fabric in production — banking, supply chain, healthcare consortia — who need to modernize aging chaincode without risking network stability, and who need an audit trail stronger than a commit log. Interaction: a developer asks Bob, in FabricPilot mode, to migrate a piece of legacy Fabric code. Bob analyzes the repo, explains what it found and why it matters, tests the migration in the sandbox, and only writes the audit record once it's proven safe — narrating every step so the developer understands both the change and the risk before it ships. What makes it unique: unlike commit-provenance tools that attach a cryptographic signature to code and store it off-chain, FabricPilot's Ledger Stamp is a genuine, endorsed Fabric transaction. Verifying it doesn't mean trusting a signature — it means asking the blockchain, which is the one thing a Fabric network is already built to answer honestly. Combined with the sandbox-first workflow, FabricPilot treats "prove it before you ship it" as a hard requirement, not a suggestion — and it adds a further IBM-native touch: the plain-language summary written into every Ledger Stamp is generated by IBM watsonx.ai (Granite), so the audit trail is both cryptographically real and human-readable from the start. The demo runs against a local Fabric network by design, since exposing live blockchain infrastructure publicly requires network hardening beyond the scope of this build.