What it does
Test Coverage Gap Agent solves the everyday problem of padding a coverage number with meaningless getter/setter tests. You give it a coverage report — per-file covered and total line counts plus each file's business-criticality tags, from Jest, Vitest or PHPUnit output — and it blends the coverage gap with what the code actually does. Payment, authentication and data-deletion outrank a heavily covered UI helper, so the riskiest untested paths surface first instead of the ones that are easiest to pad. Ranking is by exposure: the risk score multiplied by the number of untested lines. A half-covered payment file with 140 untested lines outranks a barely-covered helper with nine, which is the ordering a tech lead would actually defend in a review. Each step is its own tool, so you can ask for one answer without running everything. `list_critical` returns just the payment and auth files that must be tested first. `score_files` returns the per-file breakdown — coverage fraction, gap, weighted risk score and exposure. `generate_tests` writes the starter files. `run_full` is there as the orchestrated path, but it is one option rather than the only door. The generated tests are the part worth looking at closely. Each stub names the real failure scenario behind the gap — a duplicate charge on retry, an expired token, an unconfirmed delete — in the right framework for the file extension, rather than a happy-path check that merely runs. Two things the agent tells you rather than hiding. The projected coverage delta is labelled as an optimistic assumption, not a measurement: it assumes every generated test fully closes its file's gap. And it reminds the reviewer that a passing suite is not the goal — a low mutation score means the new tests do not really catch regressions. Every number is code-derived. The language model only narrates the summary, with a deterministic template fallback when no model is available. There are no connectors and no credentials: it runs entirely on the report you provide. Built for developers, tech leads and QA engineers.
Example prompts
- Here is our Vitest coverage report — which files should we test first?
- Show me only the critical untested payment and auth code
- Generate starter tests for the riskiest gaps
