docs: map agentic ai systems research

docs/receipts/agentic-ai-hitchhiker-research-map-20260702.json

@@ -0,0 +1,63 @@
+{
+  "schema": "fable.research_map.receipt.v1",
+  "createdAt": "2026-07-02T00:00:00.000Z",
+  "taskLabel": "agentic-ai-hitchhiker-research-map",
+  "task": "Map arXiv:2606.24937v1 into Fable's receipt-first agentic AI posture",
+  "source": {
+    "url": "https://arxiv.org/abs/2606.24937",
+    "pdfUrl": "https://arxiv.org/pdf/2606.24937",
+    "arxivId": "2606.24937",
+    "title": "The Hitchhiker's Guide to Agentic AI: From Foundations to Systems",
+    "authors": ["Haggai Roitman"],
+    "submitted": "2026-06-22",
+    "subjects": ["cs.AI", "cs.CL", "cs.IR", "cs.LG"],
+    "verifiedFrom": "arXiv abstract page metadata fetched during session"
+  },
+  "constraints": [
+    "external arXiv paper claims are advisory until locally reproduced or mapped into committed Fable receipts",
+    "agentic AI production/deployment discussion does not authorize Fable deploys",
+    "live receipt > committed receipt > memory",
+    "8099/deploy via git-proxy is the canonical signed/gated deploy route only",
+    "8098/deploy-webhook remains legacy/internal/stale for fable-agent trust",
+    "auto_patch_proven=false for fable-agent unless a fresh local degraded-skill recovery receipt proves otherwise",
+    "no deploy, external code execution, vendoring, dashboards, or new packages are included in this map"
+  ],
+  "patterns": [
+    {
+      "name": "full-stack agentic systems map",
+      "adaptation": "Use as an advisory checklist for Fable's layers: model substrate, reasoning/alignment, RAG/memory, harness, tool use, multi-agent coordination, evaluation, and production gates."
+    },
+    {
+      "name": "agent harness and context management",
+      "adaptation": "Map only into local receipts/tests before claiming Fable capability. Agents draft; receipts prove; humans approve guarded deploy."
+    },
+    {
+      "name": "agent evaluation methodology",
+      "adaptation": "Prefer denominator-rich eval receipts with attempted, confirmed, blocked, and unknown outcomes over model-written verification claims."
+    },
+    {
+      "name": "MCP/A2A/tool coordination",
+      "adaptation": "Treat protocols as integration surfaces, not authority surfaces; CLI and receipts remain the trust boundary."
+    }
+  ],
+  "blocked": [
+    "treating the paper as Fable-verified production proof",
+    "using paper deployment sections as deploy authorization",
+    "reviving stale 8098/deploy-webhook or memory-only deploy authority",
+    "claiming auto-patch/self-improvement is proven for fable-agent from memory or external paper text",
+    "vendoring code or examples without separate local review"
+  ],
+  "recommendedPlan": [
+    {
+      "name": "Agentic system layer checklist receipt",
+      "goal": "Create a local checklist mapping Fable's actual receipts to each agentic-system layer before claiming coverage.",
+      "acceptanceCriteria": [
+        "each claimed layer cites a committed or live Fable receipt",
+        "missing layers are marked unverified rather than inferred from the paper",
+        "deploy sections cite only 8099 signed/gated route receipts"
+      ]
+    }
+  ],
+  "decision": "advisory",
+  "deployAttempted": false
+}