0.2.0

Added

• Operator plugin API: define custom operators by subclassing
  ActiveMutator::Operators::Base (subclassing is registration), loaded
  via the repeatable --operator FILE flag or the operators: config-file
  key. Relative paths resolve against the project root. See
  docs/guides/custom-operators.md. The Base helpers (edit,
  loc_range) and Edit members are semver-stable from this release.
• Heredoc mutation: plain heredocs with a nonempty (dedented) body get
  an "empty heredoc body" mutant. Interpolated heredocs are untouched.
• class << self mutation: defs inside class << self within a
  constant scope are now discovered as singleton subjects and re-inserted
  through the singleton class. class << obj and top-level class << self
  stay skipped, as do classes/modules declared inside class << self
  (their constants live on the singleton class and are unreachable by
  name).
• Nested def mutation: bodies of defs nested inside another def are
  mutated under the outer subject (a nested def re-executes on every outer
  call, so a separate subject would produce phantom survivors).
• Broader CallSwap pack: all? → any?, take ↔ drop,
  min_by ↔ max_by, sort → reverse, detect/find → first.
• Project config file: .active_mutator.yml at the project root,
  layered under CLI flags (flags win).
• --fail-at SCORE: opt-in gate — exit 0 when the mutation score meets
  the threshold even with survivors.
• Adaptive timeout calibration (on by default, --no-adaptive-timeout
  to disable): grow-only budget scaling (1x–4x) from observed worker wall
  times, so parallel-load slowdown doesn't turn honest kills into false
  timeouts.
• Delta refresh now re-runs non-covering spec files that reference a
  changed file's constants, closing the newly-covering-example blind spot.
• --format stryker-json (mutation-testing-report-schema v2) and
  --format github (PR annotations for survivors).
• Per-operator equivalent-rate metric in run summaries.
• Subject expression language for --subject (Foo::Bar#baz, Foo::Bar,
  Foo::Bar*, Foo::Bar#*).
• --exclude PAT glob filtering (gitignore-like recursive semantics),
  --max-mutants N deterministic sampling, and --debug-plan dry run.
• Per-method opt-out via a # active_mutator:skip comment on the line
  above a def.
• Benchmark harness (bin/bench, bin/bench-diff) with a cross-run
  Stryker report differ.

Fixed

• Positional path arguments: nonexistent paths error out, non-Ruby file
  args are rejected, overlapping path args are deduplicated.
• Scoped --accept-survivors no longer clobbers out-of-scope ledger
  entries; accepted fingerprints referencing missing files warn.
• Covering-example lookup widened to the subject's whole line range.
• Scheduler tolerates non-Hash JSON payloads from workers.
• Unloadable operator files and operators raising during analysis fail
  with attributed, friendly errors instead of raw stack traces.

0.1.1

Fixed

• Orphan watchdog: if the parent process is killed in a way it cannot trap (SIGKILL, closed terminal, CI teardown), the scheduler detects the reparenting, kills all running workers, and aborts. Previously an orphaned run kept forking through the whole mutant queue unsupervised, which could exhaust the machine's CPU.
• The .active_mutator/ cache directory now writes its own .gitignore on creation, so host projects can never commit the disposable coverage cache by accident.

Full changelog: https://github.com/drj613/active_mutator/blob/main/CHANGELOG.md