Test Gap Finder

Line coverage overstates quality — a test can execute a line without asserting on its result, leaving a mutation-survivor that proves the behavior is untested. Combining the two datasets reveals the gap: high execution coverage plus high mutation survival equals 'we run this code but don't check the result.' That is the highest-ROI gap to close.

1. Ingest coverage report (lcov, cobertura, or similar) and mutation report (stryker, mutmut, pitest). 2. For each file, compute per-line (executed, mutant_killed, mutant_survived). 3. Rank lines by (high-execute, low-kill) — those are 'covered but untested.' 4. Cluster nearby surviving mutants into 'behaviors' using AST proximity. 5. For each behavior, prompt the LLM for a plain-English description of what the test would assert, using the surrounding code as context. 6. Emit a prioritized markdown report grouped by file, ordered by mutant survival count.