What problem does it solve? Static audits and offline test suites can only confirm a Claude Code plugin agrees with itself; they cannot detect where the plugin disagrees with the real world — live MCP server schemas, actual filesystem behavior, rendered artifacts, or human answers. This Skill finds that defect class by executing the Senzing Bootcamp plugin rather than reading it. ## Core Features & Use Cases - Phase 1 — MCP call contracts: Verifies every documented MCP call against the live Senzing MCP server's schemas, re-asks dated negative claims, and cross-checks the server's confabulation list. - Phase 2 — Hooks and scripts: Executes every hook entry and bundled script against a realistic scaffolded bootcamp project, asserting on artifacts and exit codes rather than exit status alone. - Phase 3 — Conversational layer: Walks the bootcamp with the maintainer answering as the Bootcamper, starting analysis at a chosen module to reach later modules before context runs out. - Coverage reports: Ships coverage_reports.py with invariants, affected, negatives, unmarked, and shipped reports that point audits at unguarded conformance gaps. - Use Case: A maintainer preparing a release runs all three phases to confirm the plugin actually works — catching defects like a mapping_workflow call missing its required workspace_dir parameter that three audits and 399 passing tests missed. ## Quick Start Ask the assistant to dry-run the Senzing Bootcamp plugin and specify which of the three phases to run.
