Catalog split: federated build, eval enrichment fix, UX improvements

AGENTS.md

@@ -40,6 +40,8 @@ agentic-collections-catalog/
 
 The clones are ephemeral (deleted after each build). Nothing from external repos is committed here.
 
+> **Eval data lives in source repos, not here.** `eval_site_enrichment.py` reads `eval/<pack>/<skill>/report.json` from the *cloned* source repo (e.g. `agentic-collections-skills/eval/rh-sre/remediation/report.json`). There is no `eval/` directory in this catalog repo — looking for eval files here will always find nothing.
+
 ## Marketplace File
 
 `marketplace/rh-agentic-collection.yml` is the **only** place that controls which packs appear on the site. Each module entry carries:

README.md

@@ -4,6 +4,8 @@ Unified marketplace and website for Red Hat agentic skill collections. This repo
 
 **This is not a skills development repository.** Skills are authored in source repositories like [agentic-collections-skills](https://github.com/RHEcosystemAppEng/agentic-collections-skills). An internal process fetches, evaluates, and assembles the catalog automatically.
 
+> **Note:** Evaluation reports (`eval/<pack>/<skill>/report.json`) live in the **source skills repos**, not here. This catalog repo contains no `eval/` directory — eval data is read from the temporary clones at build time.
+
 [![License](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](LICENSE)
 [![Validate Catalog](https://github.com/RHEcosystemAppEng/agentic-collections-catalog/actions/workflows/validate.yml/badge.svg)](https://github.com/RHEcosystemAppEng/agentic-collections-catalog/actions/workflows/validate.yml)
 [![Deploy GitHub Pages](https://github.com/RHEcosystemAppEng/agentic-collections-catalog/actions/workflows/deploy-pages.yml/badge.svg)](https://github.com/RHEcosystemAppEng/agentic-collections-catalog/actions/workflows/deploy-pages.yml)

scripts/build_website.py

@@ -11,7 +11,6 @@
 
 # Import our data generators
 from catalog_site_bundle import bundle_catalog_for_site
-from eval_site_enrichment import apply_eval_enrichment
 from generate_collection_pages import generate_collection_pages
 from generate_pack_data import generate_pack_data
 from generate_mcp_data import generate_mcp_data
@@ -44,10 +43,6 @@ def build_website():
     # Keep pack cards deterministic and alphabetically ordered.
     pack_data = sorted(pack_data, key=lambda p: p['name'])
 
-    print("📊 Enriching packs with eval/reports (eval/<pack>/<skill>/report.json)...")
-    apply_eval_enrichment(pack_data, root)
-    print()
-
     # Generate MCP server data
     print("🔌 Parsing MCP servers...")
     mcp_data = generate_mcp_data(pack_data)

scripts/eval_site_enrichment.py

@@ -299,14 +299,21 @@ def apply_eval_enrichment(packs: List[Dict[str, Any]], root: Path) -> None:
     """
     Mutate each pack: attach skill['evaluation'] for catalog-listed skills when
     eval/<pack>/<skill>/report.json exists; set pack['evaluation_summary'] rollup.
+
+    Eval reports live in the SOURCE repo (read from the clone at build time), not in
+    the catalog repo. The root argument must point to the cloned source repo root.
+
+    Falls back to pack['skills'] (from SKILL.md frontmatter) when no catalog is present.
     """
     root = root.resolve()
     for pack in packs:
         coll = pack.get("collection")
-        if not isinstance(coll, dict):
-            continue
+        if isinstance(coll, dict):
+            skills_ref = _iter_catalog_skills(coll)
+        else:
+            # No catalog: use raw skills parsed from SKILL.md frontmatter
+            skills_ref = [s for s in (pack.get("skills") or []) if isinstance(s, dict)]
 
-        skills_ref = _iter_catalog_skills(coll)
         catalog_skill_count = len(skills_ref)
         attached: List[Optional[Dict[str, Any]]] = []