@inflexa-ai/harness 0.6.0 → 0.7.1

package/dist/reference-data/catalog.d.ts

@@ -6,6 +6,14 @@ export declare const REFERENCE_DATA_CATALOG_VERSION: 1;
 export declare const ReferenceArtifactPathSchema: z.ZodString;
 /** Lowercase SHA-256 digest of one artifact's bytes. */
 export declare const ReferenceSha256Schema: z.ZodString;
+/**
+ * Logical content format, independent of compression: a `.txt.gz` mapping table
+ * is `tsv`, not `gz`. Agents pick a reader from this rather than guessing from
+ * the extension, so it stays a free token instead of an enum — a new source must
+ * remain cheap to add, and an unrecognized format degrades to "unknown reader"
+ * rather than failing catalog validation.
+ */
+export declare const ReferenceArtifactFormatSchema: z.ZodString;
 /**
  * One final file distributed for a reference dataset: a stable install-relative
  * path and the third-party https URL it is fetched from. The catalog carries no
@@ -15,17 +23,32 @@ export declare const ReferenceSha256Schema: z.ZodString;
  * one uniform trust-on-first-use model; there is deliberately no per-artifact
  * integrity class, so adding a source is only ever a URL, and no checked-in
  * digest is ours to maintain or lets go stale when a `current` upstream rebuilds.
+ *
+ * `format` and `contents` exist because skills name no paths: an agent asked for
+ * "a TF-target regulon network" can only match that against an inventory entry if
+ * the entry says what it holds and how to read it. `contents` should name the
+ * shape a caller must know — key columns, identifier space — not restate the title.
  */
 export declare const ReferenceArtifactSchema: z.ZodObject<{
     path: z.ZodString;
     url: z.ZodURL;
+    format: z.ZodString;
+    contents: z.ZodString;
 }, z.core.$strict>;
+/**
+ * Common-name organism the dataset describes, lowercase (`human`, `mouse`, `rat`).
+ * Omitted for species-agnostic or multi-species sources. This is the axis a wrong
+ * choice is most silently wrong on — human regulons over mouse counts still run —
+ * so it is a first-class field rather than something inferred from the id suffix.
+ */
+export declare const ReferenceOrganismSchema: z.ZodString;
 /** Provenance and licensing information for one supported reference dataset. */
 export declare const ReferenceDatasetSchema: z.ZodObject<{
     id: z.ZodString;
     version: z.ZodString;
     title: z.ZodString;
     description: z.ZodString;
+    organism: z.ZodOptional<z.ZodString>;
     sourceUrl: z.ZodURL;
     license: z.ZodObject<{
         identifier: z.ZodString;
@@ -38,6 +61,8 @@ export declare const ReferenceDatasetSchema: z.ZodObject<{
     artifacts: z.ZodArray<z.ZodObject<{
         path: z.ZodString;
         url: z.ZodURL;
+        format: z.ZodString;
+        contents: z.ZodString;
     }, z.core.$strict>>;
 }, z.core.$strip>;
 /** Versioned canonical catalog consumed identically by every harness embedder. */
@@ -48,6 +73,7 @@ export declare const ReferenceDataCatalogSchema: z.ZodObject<{
         version: z.ZodString;
         title: z.ZodString;
         description: z.ZodString;
+        organism: z.ZodOptional<z.ZodString>;
         sourceUrl: z.ZodURL;
         license: z.ZodObject<{
             identifier: z.ZodString;
@@ -60,6 +86,8 @@ export declare const ReferenceDataCatalogSchema: z.ZodObject<{
         artifacts: z.ZodArray<z.ZodObject<{
             path: z.ZodString;
             url: z.ZodURL;
+            format: z.ZodString;
+            contents: z.ZodString;
         }, z.core.$strict>>;
     }, z.core.$strip>>;
 }, z.core.$strip>;
@@ -83,9 +111,11 @@ export declare function referenceArtifactKey(dataset: {
 /**
  * Canonical release catalog. Every artifact is fetched directly from the third
  * party that publishes it; this project hosts, mirrors, and redistributes
- * nothing. Entries are added with a real upstream https URL, provenance, and
- * licensing data — and nothing else per file: no size, no digest to compute or
- * keep in sync.
+ * nothing. Entries are added with a real upstream https URL, provenance and
+ * licensing data, and a description of what the file holds — but never a size or
+ * digest to compute and keep in sync. The descriptive half is what makes a
+ * dataset findable: this catalog is the only place that says what a reference
+ * file is, so nothing downstream has to encode a path or a filename to use it.
  *
  * `version` is the dataset's upstream release identifier. Datasets built on an
  * upstream that has no immutable release — NCBI regenerates `gene_info` daily,