[e54-bot]

What changed

Extends the SPEC-309 declarative lint matcher so the four structural shapes
behind the built-in rules become expressible (#970, owner decisions
2026-10-27 applied: declarative finding format, ongoing-condition-hot-path
stays native, loop-scope values searches body actions only).

• matcher.scope accepts a list of scopes — one rule ID enumerates all
  listed scopes ([while, for-global-variable, for-player-variable] shares a
  rule). A single scope behaves as before.
• matcher.values lists value patterns searched at any depth in every
  value tree of the scope: rule conditions + live action arguments for
  rule; live body-action arguments for a loop scope (not the loop's own
  condition; a nested loop's body is searched once per enclosing loop); the
  condition list for if. Disabled/dead statements stay excluded.
• matcher.report: each-match anchors one finding per matched node at the
  node's own span; default scope is unchanged.
• A values pattern may carry within (+ in-args: { from: N }), an
  ancestor constraint limited to ancestor argument positions ≥ N — so
  Filtered Array(Sorted Array(x, r), c) is not reported.
• A call pattern accepts names: [...] as an alternative to name; every
  spelling resolves through the catalog at load time.
• First-party package rules may declare evidence and severity in
  metadata (Support GitHub-based community lint rule discovery and lifecycle management #616 addendum); both fields are rejected with lint-rule-error
  in community packages and --rule-file rules.
• declarative.rs split into declarative/{schema,pattern,error}.rs
  (size-guard), SPEC-309 + ADR-0015 amended, and the
  reproducer-bastion-prophet.ws fixture added (verbatim copy from
  workshop-rs import-limits, SHA-256 3a1cb891…, Apache-2.0).

Verification

• crates/wright-analyzer/tests/yaml_contract.rs gains cases for every
  acceptance criterion: scope lists, each-match per-node anchoring at any
  depth with disabled exclusion, within/in-args binder semantics,
  names resolution + unknown-spelling rejection, first-party evidence/
  severity acceptance vs. external rejection, and span-parity checks of
  nested-binder-lambda/expensive-loop-check/min-wait-loop written in
  the new vocabulary against the native rules on
  reproducer-bastion-prophet.ws (16 findings — see note) and on lint
  fixtures.
• cargo fmt --all -- --check: clean.
• cargo clippy --workspace --all-targets --all-features -- -D warnings: clean.
• cargo test --workspace --all-targets --all-features: passing (see run
  notes below).
• git diff --check: clean.

Unverified / notes

• The issue body says the reproducer should yield 13 nested-binder
  findings; the native rule is the acceptance reference and reports 16
  on the current fixture, matching the count workshop-rs
  docs/import-limits.md records for this file. Span parity holds; the test
  asserts parity + 16.
• expensive-loop-check's native target list is distance, raycast,
  isInLoS — raycast is not a workshop-rs catalog id, so the declarative
  fixture lists the two resolvable spellings (distance, isInLoS), which
  is the effective native set; a YAML names: [raycast] would be a load
  error by design (unknown spelling).

Refs #970

[e54-bot]

Review finding addressed:

Finding 1 — each-match findings under a non-rule scope addressed the owning statement instead of the scope anchor, diverging from the native loop rules on the machine-readable action field.

Fixed in 849418e: the each-match arm in run_canonical now resolves action as anchor_action.or(owner_action), so a loop scope reports the loop (matching native expensive-loop-check) while a rule scope keeps the owning statement (matching native nested-binder-lambda). The span-parity tests now key on (span, action) so the addressed node is asserted.

Verification: cargo test --workspace --all-targets --all-features (65 suites, 0 failures), cargo clippy --workspace --all-targets --all-features -- -D warnings clean, cargo fmt --all -- --check clean, git diff --check clean.

Ready for follow-up review.

[Teakowa]

LGTM after independent review.