v1.1.0

• Support Psalm 7: TKeyedArray's constructor is private there, so shapes are
  built through TKeyedArray::make() (which Psalm 6 has from 6.19). The
  vimeo/psalm constraint widens to ^6.19 || ^7.0. The Returns integration
  fixture now accepts InvalidScalarArgument, which is what Psalm 7 calls the
  wrong-callback-return case Psalm 6 filed as InvalidArgument.

v1.0.0

The stability release, following rasuvaeff/understudy 1.0.0. The plugin's
diagnostic contract is unchanged from 0.8.5; a project on the 0.x bridge
upgrades by changing its engine constraint to ^1.0.

• Requires rasuvaeff/understudy ^1.0, following the family policy: the
  engine is released first and satellites follow it on the same major.

v0.8.5

• Changed. Accepts rasuvaeff/understudy ^0.12 alongside ^0.9 || ^0.10 || ^0.11: the engine's 0.12.0 runs a specification closure exactly once (the nested-call probe that re-ran it is gone) and refuses two calls side by side as it does a nested pair; this package reads neither, so the bridge is widened, not moved.

v0.8.4

• Changed. Accepts rasuvaeff/understudy ^0.11 alongside ^0.9 || ^0.10: the engine's 0.11.0 adds a refusal where there was a fatal, a nested-call check and a returns() type check, none of which this package reads; the bridge is widened, not moved.

v0.8.3

• Requires rasuvaeff/understudy ^0.9 || ^0.10. The engine's 0.10 closes six
  defects of the 1.0 review, the widest of them being that an optional
  parameter no longer has to be spelled in a specification: when(fn () => $repository->find(1)) now matches a contract whose remaining parameters
  carry defaults, and an unspelled position is rendered as … in reports so
  it is never mistaken for a written any(). Nothing this plugin does changes
  — the arity suppression it carries is about Arg::rest(), which still
  covers the contract's required parameters and only those.
• Drops the ^0.8 term, as 0.8.2 said it would in the release that
  follows the engine's.

v0.8.2

• Guards the plugin's load list with a regression test and completes the
  plugin's public summary in the generated family reference.
• Removes the stale 1.0 wording from llms.txt while the engine family remains
  on its 0.x compatibility line.

v0.8.1

• Allows rasuvaeff/understudy ^0.8 || ^0.9. A bridge, not a widening for
  its own sake: the engine's 0.9 is the 1.0 candidate — every contract decision
  of the 1.0 review lands there, and 1.0 follows once it has been driven by real
  packages — and a project taking it must be able to keep this plugin without a
  window in which composer require silently installs the 0.8 engine beside
  it. The ^0.8 term is dropped in the release that follows the engine's.
• Rector is green again, which it had not been since 0.5.0: rector.php
  skips the fixture projects — they are the test input, and
  SortCallLikeNamedArgsRector would have rewritten the Misuse fixture
  that pins named bounds in either order into the order that never showed the
  bug — and the remaining suggestions are applied. composer release-check
  is what gates a tag, and it had been red under green builds.
• infection/infection moves to ^0.35, the monorepo's single-major form.
• Documentation no longer dates the Arg::rest() and Arg::captor() idioms to
  «understudy 0.4»: with a floor of ^0.8 every engine this plugin installs
  beside has them. AGENTS.md no longer cites a section of the retired plan.

v0.8.0

• Requires rasuvaeff/understudy ^0.8, and requires it as a single term. The
  accumulating union it carried (^0.4 || ^0.5 || …) had to be widened by hand
  on every core release, and a package that misses one becomes uninstallable
  beside its own engine.
• The plugin no longer crashes Psalm on foo(...). A first-class callable
  carries a VariadicPlaceholder where its arguments would be, and php-parser
  asserts against reading them; the before hook read them on every call it
  saw, so a single strlen(...) anywhere in the analysed code took the whole
  run down — no diagnostics, on any file, at any level. That is PHP 8.1 syntax
  and this package declares 8.3–8.5.
• ->returns(...)->times(5, 2) is reported as impossible again. The
  cardinality rule read only the immediate receiver of times(), so it fired
  on expect(...)->times(5, 2) and stayed silent on every spelling with a
  link in between — including the one the engine's own README recommends for a
  repeated call.
• times(maximum: 5, minimum: 1) is no longer reported as impossible. The
  bounds were read positionally, so a named call with the arguments in the
  other order looked like (5, 1). Correct code turned red, with no way around
  it but removing the plugin.
• A specification that reaches its double through a helper is silent.
  when(fn () => $this->gate()->find(1)), $double->find($this->id()) and
  $this->passThrough($double->find(1)) were all reported as "the closure
  makes 2 calls". The engine throws on the first call that lands on a double
  and never sees the rest, so a call that is the receiver of another, or one
  made on $this, is not a second specified call. The total is still what
  answers "no call at all", so a specification that only reaches its double
  through a helper is not accused of specifying nothing either.
• A matcher whose kind cannot fit is described as matching "a value of type
  int" rather than "a int" — the wording understudy-phpstan already used.
• README ×2 and llms.txt now say that a leaked matcher is reported by Psalm's
  own MixedArgument and therefore only at errorLevel="1", and why this
  plugin has no rule of its own for it.

v0.7.0

• Matcher argument suppression is now scoped to real matcher calls.
  Arg::captor() builds a captor and is no longer mistaken for a matcher, so
  Psalm keeps reporting an invalid captor value passed directly to a typed
  parameter.
• TooFewArguments suppression now uses file offsets for every scope. A
  valid Arg::rest() specification and a real under-arity call on the same
  source line are distinguished, so the real call keeps Psalm's diagnostics.
• Added control-run integration coverage for both boundaries.

v0.6.1

• Documentation review fixes. Added the missing make test-integration
  target — README and AGENTS referenced it, the Makefile did not define it.
  Both READMEs gained the Security and Examples sections; the make build
  comment now lists everything the script runs.