What problem does it solve? When an MCP server's tool surface changes or its documentation drifts from the live contract, there is no systematic way to verify that every exported tool actually works as documented. This Skill provides a disciplined expectations-first audit method so tool testing produces verifiable findings instead of exploratory guessing. ## Core Features & Use Cases - Expectations-first audit: Write an expectations table (tool, expected behaviour, expected response shape) from docs, live schemas, and mcp-index intents before making any call, then compare actual responses against it. - Destructive-tool safety controls: Use a dedicated test project, positive and negative controls for delete/sweep/share tools, and verified zero-residue cleanup. - Dependency-ordered execution with verdicts: Batch independent calls, sequence dependent ones, and record PASS/PARTIAL/FAIL verdicts plus a findings section covering docs drift and response-shape inconsistencies. - Use Case: A server's surface grew from 19 to 21 tools. Run the full surface test, commit a results doc to the repo showing 33 PASS / 2 PARTIAL, and fold newly discovered quirks into the server's pitfalls documentation. ## Quick Start Test every tool exported by the memory MCP server and commit a results doc to the repo's work-docs directory.
