Test · Configuration

The Test family is opt-in. Off by default — turn it on once your project has (or is willing to wire up) an lcov.info produced by cargo llvm-cov, pytest --cov, nyc, or scoverage. heal never executes tests itself; everything that needs the test suite to actually run (flakiness, mutation score, runtime trends) stays out of scope.

For what each metric flags, see Test › Metrics. For the bundled skills, see Test › Skills.

Quick enable

[features.test]
enabled = true

[features.test.coverage]
enabled = true

Defaults cover Rust / TypeScript / JavaScript / Python / Go / Scala test conventions and the four conventional lcov.info paths. Most projects don’t need to override anything.

If you don’t have an lcov.info yet, run the bundled setup skill — it inspects your stack and proposes the reporter wiring.

claude /heal-test-reporter-setup

For the full skill contract see Test › Skills.

`[features.test]`

[features.test]
enabled    = false                # master switch
test_paths = [
  "tests/**",
  "**/*_test.rs",
  "**/*.test.ts", "**/*.test.tsx", "**/*.test.js", "**/*.test.jsx",
  "**/*.spec.ts", "**/*.spec.tsx", "**/*.spec.js", "**/*.spec.jsx",
  "**/__tests__/**",
  "**/*_test.go",
  "**/test_*.py", "**/*_test.py",
  "**/*Test.scala", "**/*Spec.scala",
]

enabled (default false) — master switch. While false, every test observer is a no-op.
test_paths (default: language conventions above) — gitignore- syntax globs that mark which source files are tests. The skip_ratio observer walks these files; every Finding whose primary file matches is also tagged is_test_file = true.

When test_paths is empty, heal falls back to a built-in heuristic covering the same conventions.

`is_test_file` flag

When [features.test] is enabled, every Finding gains an is_test_file: bool flag. Skills filter on this to read test- and production-side severities independently — /heal-test-review focuses on test findings; /heal-code-review focuses on production findings.

The flag is omitted from JSON output when false, so projects that don’t enable the test family see byte-identical latest.json content to before.

`[features.test.coverage]`

[features.test.coverage]
enabled    = false
lcov_paths = [
  "lcov.info",
  "coverage/lcov.info",
  "target/llvm-cov/lcov.info",
  "coverage/lcov-report/lcov.info",
]

enabled (default false) — sub-feature switch. Keep [features.test] on but [features.test.coverage] off when you want is_test_file tagging and skip_ratio without yet wiring up a reporter.
lcov_paths — project-relative paths probed in order. First existing file wins; the rest are ignored. Missing files are silent — no warning at startup.
post_commit_refresh (default unset) — optional shell command the post-commit hook spawns in the background to re-run your reporter after every commit. The process is detached and its output discarded so your commit flow doesn’t wait. Pair it with the same command /heal-test-reporter-setup proposes (cargo llvm-cov --workspace --lcov --output-path lcov.info --locked --ignore-run-fail, pytest --cov=..., etc.) so the next heal status reads fresh lcov.info. Skipped silently when [features.test] or [features.test.coverage] is off.

heal reads what your CI / local reporter produces. The default probe order covers:

Reporter	Path written
`cargo llvm-cov --lcov`	`target/llvm-cov/lcov.info`
`pytest --cov --cov-report=lcov`	`coverage/lcov.info`
`nyc --reporter=lcov`	`coverage/lcov-report/lcov.info`
`scoverage` (Scala)	varies; symlink to `lcov.info` if needed

The lcov reader is permissive — it tolerates unknown record types and recovers totals from per-line records when reporters omit the summary fields, so most reporter dialects work out of the box.

Calibration

Two new sections appear in .heal/calibration.toml when you run heal calibrate --force with the test family on:

[calibration.coverage_pct]
# Heal stores INVERTED values (100 - coverage_pct) so the same
# `value >= p95 → Critical` cascade applies as for the rest of the
# metrics — worst still maps to Critical.
p50 = 30.0     # 70% coverage
p75 = 50.0     # 50% coverage
p90 = 70.0     # 30% coverage
p95 = 85.0     # 15% coverage
floor_critical = 95.0   # ≤ 5% coverage → Critical regardless of percentile
floor_ok       = 25.0   # > 75% coverage → Ok regardless of percentile

[calibration.skip_ratio]
p50 = 0.0
p75 = 1.0
p90 = 5.0
p95 = 10.0
floor_critical = 20.0   # > 20% skip → Critical
floor_ok       = 0.5    # < 0.5% skip → Ok

These are the literature-anchored fallbacks heal uses until you run heal calibrate --force. Floors belong in config.toml, not here, so they survive recalibration:

[metrics.coverage_pct]
floor_critical = 90.0   # tightens to "≤ 10% coverage → Critical"

[metrics.skip_ratio]
floor_ok = 0.0          # any skipped test surfaces

(coverage_pct overrides apply to the inverted form — floor_critical = 90.0 means ”≤ 10% line coverage”, not ”≤ 90%”.)

Post-commit nudge

When [features.test.coverage] is on, the post-commit hook adds an indented second line to the nudge:

heal: recorded · 3 critical, 7 high · heal status
         · 2 uncovered hotspot

The count is the number of coverage_pct findings at High or Critical severity that also carry hotspot=true. The line is suppressed when the coverage feature is off.

Strict by design

Like every other section, [features.test] and [features.test.coverage] reject unknown keys:

[features.test]
test_path = ["tests/**"]   # ✘ unknown — heal errors here
                            #   (it's `test_paths`, plural)