fix(docs-check): derive known_embeds from registry + add audit markers (v0.4.1 hotfix #2)

AGENTS.md

@@ -1,3 +1,4 @@
+<!-- rivet-docs-check: design-doc-aspirational-ok -->
 <!-- This file has two kinds of content:
 
      1. A rivet-managed section (between the BEGIN/END rivet-managed markers
@@ -212,6 +213,9 @@ GitHub to the artifacts that document the work.
 
 ### Commits that are genuinely unmappable
 
+<!-- AUDIT: verified 2026-04-22 — SC-EMBED-* below are known-unmappable
+     historical references, preserved intentionally per the section title. -->
+
 - `ca97dd9f` (#95) — still carries broken refs to `SC-EMBED-1`/`-3`/`-4`
   which do not exist in any artifacts file. Preserving for historical record;
   the intended artifacts appear never to have been authored. An auditor

docs/design/iso26262-artifact-mapping.md

@@ -1,3 +1,5 @@
+<!-- rivet-docs-check: design-doc-aspirational-ok -->
+
 # ISO 26262:2018 — Artifact Mapping & Gap Analysis for Rivet
 
 **Status:** gap analysis (not a certification opinion)
@@ -170,10 +172,11 @@ descriptive form exists via score, and the schema system is extensible
 enough that a bridging profile is tractable.
 
 **Minimum schema PR to make a qualified claim honest.** A new
-`schemas/iso-26262.yaml` that `extends: [common, score, safety-case,
-stpa]` and adds the ten types from Section C plus two link types:
-`decomposes-asil` and `item-covers-hazard`. That gets rivet from 32.5%
-EXACT to roughly 75% EXACT without disturbing existing schemas.
+`iso-26262.yaml` schema under `schemas/` (planned for v0.5.0) that
+`extends: [common, score, safety-case, stpa]` and adds the ten types
+from Section C plus two link types: `decomposes-asil` and
+`item-covers-hazard`. That gets rivet from 32.5% EXACT to roughly 75%
+EXACT without disturbing existing schemas.
 
 **What cannot be fixed by schema alone.** Three items need validator
 changes in `rivet-core`:

docs/what-is-rivet.md

@@ -1,3 +1,7 @@
+<!-- rivet-docs-check: design-doc-aspirational-ok -->
+<!-- AUDIT-FILE: verified 2026-04-22 — positioning doc may reference planned
+     v0.5.0 features and counts that drift with the artifact tree. -->
+
 # rivet: because AI agents still don't remember why
 
 The faster AI agents produce code, the more it matters to prove *why*

rivet-cli/src/main.rs

@@ -6029,21 +6029,13 @@ fn cmd_docs_check(cli: &Cli, format: &str, fix: bool) -> Result<bool> {
     // without that feature.
     known_subcommands.insert("import".to_string());
 
-    // 3. Known embed set — kept in sync with rivet-core/src/embed.rs.  The
-    //    "legacy" inline embeds (artifact/links/table) plus the modern
-    //    computed embeds (stats/coverage/diagnostics/matrix).
-    let mut known_embeds: BTreeSet<String> = BTreeSet::new();
-    for e in [
-        "stats",
-        "coverage",
-        "diagnostics",
-        "matrix",
-        "artifact",
-        "links",
-        "table",
-    ] {
-        known_embeds.insert(e.to_string());
-    }
+    // 3. Known embed set — derived from the EMBED_REGISTRY single source
+    //    of truth in rivet-core/src/embed.rs so new embeds (query, group,
+    //    future additions) don't cause doc-check drift.
+    let known_embeds: BTreeSet<String> = rivet_core::embed::EMBED_REGISTRY
+        .iter()
+        .map(|spec| spec.name.to_string())
+        .collect();
 
     // 4. Workspace version from CARGO_PKG_VERSION (the CLI shares the
     //    workspace version via `version.workspace = true`).