jmagly Test platform research
Read the target manifests, installed runner versions, CI commands and current conformance protocol. Run aiwg test-conformance research against configured local roots and the selected platform. Missing roots or unavailable web
access become explicit gaps. User-specific home paths are configuration, not shipped defaults.
Use tool recommendation for decisions and adapter qualification before claiming a platform works. Distinguish a researched platform recipe, a fixture-tested result parser, and a live end-to-end verified target; evidence at one level does not prove the next.
Search user-authorized research roots read-only. Prefer the user's archived paper/reference material when relevant, then verify current tool behavior using official documentation and primary publications. Record exact local paths or source URLs, retrieval date, version applicability, claims, transfer limitations and unresolved contradictions. A paper's result on one language or dataset is not a universal quality target.
Evaluate additions only against a concrete gap: mutation for weak oracles, property testing for broad input spaces, browser testing for browser boundaries, contract testing for a real service interface, linting for syntactic misuse, or repeat-run analysis for suspected flakes. Include cost, runtime, prerequisites, CI integration and what the tool cannot prove. Research recommendations do not install dependencies or run network services.
Use templates --action list, templates --action develop and templates --action deploy to adapt a platform bundle.
Supply actual source/test globs, runner argv, result/discovery mappings, coverage scope and fixture conventions;
identify any required external adapter as unimplemented until it is built and qualified. Use the canonical result format
for custom runners and validate it before assessment. Keep original raw results alongside normalized evidence.
For optional deeper mutation, flaky or factory work, discover the existing testing-quality capability and use it rather than duplicating its implementation. Return a reviewable template/protocol, dated research references, tool choices and qualification evidence with remaining gaps.