phantom-quench

What problem does it solve?

AI-generated artifacts like test cases, analysis reports, and design documents often contain claims that are not actually present in the project's source materials — these unfounded "phantom claims" come from the model's training data rather than the actual project sources, leading to incorrect tests, invalid documentation, and faulty decision-making.

Core Features & Use Cases

• Automated Claim Extraction: Pulls all verifiable claims (proper nouns, numerical values, branching conditions, state transitions) from artifacts for back-tracing.
• Source Back-Tracing: Greps declared local source files to confirm claims exist in the correct context, and fetches external citations to verify they actually support the claimed fact.
• Severity Classification & Prescriptions: Grades ungrounded claims by risk (S/A/B) and provides specific fix steps, plus detects structural generation issues like source files not being read during artifact creation.
• Use Case: A QA team using AI to generate test cases from a requirements document can run this audit to catch test steps that reference non-existent API endpoints, incorrect threshold values, or unsupported external citations before the test cases are added to the suite.

Quick Start

Use the phantom-quench skill to audit the test case file 'user_flow_tc.md' against its declared source files 'product_requirements.md' and 'api_spec_v2.yaml' to identify any unfounded claims.