Test · Skills

The opt-in Test family is served by two skills from the heal Claude Code plugin: the tests step of /heal:setup, which enables the family and wires up a coverage reporter, and /heal:tests, which works on the tests themselves. For installing the plugin, see Code › Skills.

/heal:setup tests — enable the family and wire up lcov

Turns [features.test] on with test_paths and lcov_paths fitted to your repository, then detects your language stack and sets up a coverage reporter so lcov.info lands where heal reads it.

Stack	Reporter
Rust	cargo llvm-cov --lcov --output-path lcov.info
Python	pytest --cov=src --cov-report=lcov (pytest-cov)
JS / TS	nyc --reporter=lcov mocha / vitest --coverage --coverage.reporter=lcov
Go	go test -coverprofile=coverage.out + gcov2lcov
Scala	scoverage plugin + lcov reporter
Mixed	one reporter per stack

Every step — installing the reporter, editing .heal/config.toml, running the reporter — asks first. It can also set [features.test.coverage].post_commit_refresh so the post-commit hook refreshes lcov.info after each commit. CI changes are printed as a proposal to copy, not applied. heal doctor reports the test area as todo when coverage is on but no lcov file exists, so /heal:setup offers the step again.

/heal:tests — review and improve the tests

Works like /heal:refactor, for tests. It reads the test findings, proposes changes, and applies only the ones you approve — one commit per proposal, with the test suite green before every commit. It never pushes or opens a pull request.

1. Diagnose. Reads heal status --all --feature test --json, then each flagged source file with its tests. Uncovered code is sorted by the layer that should test it, following the test pyramid: pure logic gets unit tests; code that coordinates other modules gets a small integration test; code at an I/O boundary gets a thin contract test. Skips are sorted by their reason (environment, slow, broken, pending).
2. Propose. A short reading of the suite’s shape — an inverted pyramid, hotspots without tests, skips piling up, tests that no longer follow their source — then numbered proposals, highest-priority findings first. Each test proposal names the cases it will assert.
3. Choose. You pick which proposals to apply.
4. Apply. One commit per proposal; the suite must pass, and the coverage reporter is re-run when coverage changes.
5. Report. What changed and what is left.

What it proposes, by finding:

Metric	Typical proposal
coverage_pct	Unit tests for documented pure logic; an integration or contract test for coordination and I/O code.
change_coupling.drift	Update the test to the source’s current contract, or rewrite it when the behaviour moved elsewhere.
skip_ratio	Re-enable a test whose skip reason no longer holds; delete one whose feature is gone; ask about the rest.

When the same problem repeats across files it proposes suite-level changes too: rebalancing an inverted pyramid, splitting a suite, extracting a shared fixture, or retiring a test file. Code whose intended behaviour is not documented anywhere is raised as a question — writing a test would only cement a guess.

Refusals: never weakens an assertion to make a test pass, never skips a flaky test to lower skip_ratio, never commits a test it has not run, and never lowers coverage thresholds or turns the family off to make findings go away.

With Semantic (Jev) enabled, it can also remove tests that check nothing — a test_value finding marked delete at high confidence, one test per commit, only while the suite passes and coverage does not drop (unless the test only checked its own mocks); merge or remove duplicated tests (test_duplicate); and replace mocks of plain values with the real values (mock_scope). After each commit it runs heal semantic ask --task verify_tests --diff HEAD~1..HEAD and undoes a commit whose new test only checks its own mocks.

Arguments: a path narrows the work to findings under it, a finding id narrows it to that finding, and plan stops after the proposals.

Trigger phrases: “review the test health”, “fix the test findings”, “add the tests heal flagged”, “remove useless tests”, “/heal:tests”.