Summary

Closes #190 — the issue reported that the s-expr DSL has no discoverable way to express "this artifact has at least one outbound link of type T." linked-by already does that semantically, but the issue documents that authors (the issue author included) read it as inbound and reach for spellings that don't exist (linked-via, out-link, has-link, linked-as). The four linked-* operators also have no semantic table in the user-facing docs, so users can't reverse-engineer the right form from the listing.

This PR adds the discoverable spelling, documents the family, and regenerates the parse-error hint from the lowerer's source of truth so the next added operator cannot leave the hint stale.

What changed

• rivet-core/src/sexpr_eval.rs
  • New (linked-via "T") form. Lowers to Expr::LinkedBy(T, Wildcard) so the evaluator + links-count complement remain one code path. Arity = 1.
  • HEADS extended with "linked-via".
  • classify_filter_error now builds the unknown-head hint from HEADS (sorted) rather than a hand-maintained string. The previous opaque linked-* is replaced with the explicit family.
  • 6 new unit tests + the existing parse_error_unknown_head_surfaces_note updated for the regenerated hint.
• docs/getting-started.md
  • 7-row semantic table covering linked-via, linked-by (1- and 2-arg), linked-from (1- and 2-arg), linked-to, links-count — direction, filter axis, truth condition. Plus a mnemonic paragraph.
  • 3 gap-hunt examples: AS missing outbound exploits, REQ missing inbound verifies, hazard missing inbound prevents.
  • Predicate listing and the linked-* cross-reference both extended with linked-via.
• rivet-core/tests/sexpr_doc_examples.rs
  • 2 new doc-example regression tests: (linked-via "T") ≡ (linked-by "T" _) and the gap-hunt (and (= type …) (not (linked-via "T"))).
  • docs_listed_predicates_all_parse_as_forms extended with (linked-via "satisfies").

Acceptance — per the 2026-04-26 triage comment on #190

• New s-expr predicate (linked-via "<link-type>") returns true iff the artifact has at least one outbound link of that type. Tests filter_linked_via_outbound_present / filter_linked_via_outbound_absent / filter_linked_via_arity in sexpr_eval.rs; (linked-via "T") ≡ (linked-by "T" _) proven by filter_linked_via_equivalent_to_linked_by_wildcard and docs_example_linked_via_outbound_membership.
• Existing operator family (linked-from, linked-to, linked-by, linked-via) documented in docs/getting-started.md with a one-line semantic per operator and a worked example. New "Link predicates: which one do I want?" section + 3 worked gap-hunt examples + mnemonic paragraph.
• Error message in rivet query --sexpr lists the full operator family, regenerated from a single source of truth. classify_filter_error now emits HEADS.sort_unstable(); pinned by the regression test unknown_head_hint_lists_linked_via.
• Tests cover: artifact with the link-type → present; artifact without it → absent; (not (linked-via "X")) finds gaps. filter_linked_via_outbound_present, filter_linked_via_outbound_absent, filter_not_linked_via_finds_gap, docs_example_not_linked_via_finds_gap.
• Snapshot/doctest covers each operator's documented semantics. docs_listed_predicates_all_parse_as_forms now exercises linked-via; the two new doc-example tests assert (linked-via …) against the getting-started.md examples directly.
• Commit trailer Implements: REQ-007 (CLI/query). Trailer present; also Verifies: REQ-007 for the test additions.

Test plan

• cargo test -p rivet-core --lib — 932 pass (+6 sexpr cases)
• cargo test -p rivet-core --test sexpr_doc_examples — 11 pass (+2 cases)
• cargo test -p rivet-cli — pass
• cargo clippy --workspace --all-targets -- -D warnings — clean
• cargo fmt --all -- --check — clean
• cargo run --release -p rivet-cli -- validate — 6 errors / 140 warnings / 0 broken cross-refs — byte-identical to pristine main; the 6 errors live in the spar external fixture and are unaffected.
• CI on this PR — verified by GHA.

Out of scope

The issue body's section #2 ("linked-* operator family needs documentation") is fully addressed by the new section in getting-started.md; the optional split into a dedicated docs/sexpr.md (mentioned in the triage comment as an alternative) is not done — the existing s-expr filtering section already lives in getting-started.md and adding a parallel page would duplicate. Happy to lift it out as a follow-up if reviewers prefer.

🤖 Generated with Claude Code — issue-triage agent run 2026-05-09.

Generated by Claude Code