Agent Skills Pipeline v1.0

8-stage pipeline: Scout → Filter → Reader → Extractor → Score → Generator → Reviewer → Publisher
- Scout: GitHub search with token auth + rate limit retry
- Filter: Deterministic rules (language, stars, age, keywords)
- Reader: Incremental context loading (README → docs → examples → code)
- Extractor: LLM workflow extraction with JSON retry
- Score: Rule-based evaluation (no LLM)
- Generator: Standardized Hermes Skill format
- Reviewer: Independent LLM review (separate from generator)
- Publisher: Branch + PR to Gitea

First run: 5 repos discovered, 0 extracted (correct — all frameworks, no workflows)

runs/20260805-053642.json

@@ -0,0 +1,21 @@
+{
+  "run_id": "20260805-053642",
+  "started_at": "2026-08-05T05:36:42.627714",
+  "stages": {
+    "scout": {
+      "count": 2
+    },
+    "filter": {
+      "kept": 2,
+      "rejected": 0
+    }
+  },
+  "results": {
+    "extracted": 0,
+    "scored": 0,
+    "generated": 0,
+    "reviewed": 0,
+    "published": 0
+  },
+  "ended_at": "2026-08-05T05:36:49.213968"
+}