diff --git a/src/forge/integrations/jira/models.py b/src/forge/integrations/jira/models.py
index 836fc3fd..b32c406c 100644
--- a/src/forge/integrations/jira/models.py
+++ b/src/forge/integrations/jira/models.py
@@ -119,20 +119,71 @@ def _extract_text_from_adf(adf: dict[str, Any]) -> str:
         if not isinstance(adf, dict):
             return str(adf) if adf else ""
 
-        content = adf.get("content", [])
-        texts = []
-
-        for node in content:
-            if node.get("type") == "paragraph":
-                para_texts = []
-                for child in node.get("content", []):
-                    if child.get("type") == "text":
-                        para_texts.append(child.get("text", ""))
-                texts.append("".join(para_texts))
-            elif node.get("type") == "text":
-                texts.append(node.get("text", ""))
-
-        return "\n\n".join(texts)
+        # Jira does not provide a supported ADF-to-Markdown endpoint. Keep this
+        # extractor local and conservative so workflow prompts receive readable
+        # issue text without taking a dependency on an under-supported converter.
+        # Add unsupported ADF nodes here with focused JiraIssue regression tests.
+        # If a node cannot be parsed, preserve its raw representation instead of
+        # dropping issue content from workflow prompts.
+        def inline_text(nodes: list[dict[str, Any]]) -> str:
+            parts = []
+            for child in nodes:
+                child_type = child.get("type")
+                if child_type == "text":
+                    parts.append(child.get("text", ""))
+                elif child_type == "hardBreak":
+                    parts.append("\n")
+                else:
+                    parts.append("\n".join(extract_blocks(child)))
+            return "".join(parts)
+
+        def extract_blocks(node: dict[str, Any]) -> list[str]:
+            node_type = node.get("type")
+            content = node.get("content", [])
+
+            if node_type == "doc":
+                return extract_children(content)
+            if node_type == "paragraph":
+                text = inline_text(content).strip()
+                return [text] if text else []
+            if node_type == "heading":
+                text = inline_text(content).strip()
+                if not text:
+                    return []
+                level = node.get("attrs", {}).get("level", 1)
+                return [f"{'#' * int(level)} {text}"]
+            if node_type == "text":
+                text = node.get("text", "")
+                return [text] if text else []
+            if node_type == "codeBlock":
+                text = inline_text(content)
+                language = node.get("attrs", {}).get("language", "")
+                fence = f"```{language}".rstrip()
+                return [f"{fence}\n{text}\n```"]
+            if node_type in ("bulletList", "orderedList"):
+                items = []
+                for index, item in enumerate(content, start=1):
+                    item_text = "\n".join(extract_blocks(item)).strip()
+                    if item_text:
+                        prefix = f"{index}. " if node_type == "orderedList" else "- "
+                        items.append(prefix + item_text.replace("\n", "\n  "))
+                return items
+            if node_type == "listItem":
+                return extract_children(content)
+            if node_type == "rule":
+                return ["---"]
+
+            blocks = extract_children(content)
+            return blocks or [str(node)]
+
+        def extract_children(nodes: list[dict[str, Any]]) -> list[str]:
+            blocks = []
+            for child in nodes:
+                blocks.extend(extract_blocks(child))
+            return [block for block in blocks if block]
+
+        blocks = extract_blocks(adf)
+        return "\n\n".join(blocks) if blocks else str(adf)
 
 
 @dataclass
diff --git a/tests/unit/integrations/jira/test_client.py b/tests/unit/integrations/jira/test_client.py
index ad2977c2..0b011f9b 100644
--- a/tests/unit/integrations/jira/test_client.py
+++ b/tests/unit/integrations/jira/test_client.py
@@ -6,6 +6,7 @@
 import pytest
 
 from forge.integrations.jira.client import JiraClient, MissingProjectConfig
+from forge.integrations.jira.models import JiraIssue
 from forge.models.workflow import ForgeLabel
 
 
@@ -365,6 +366,65 @@ def test_text_to_adf_inline_formatting(self):
         # Should have paragraph with inline marks
         assert adf["content"][0]["type"] == "paragraph"
 
+    def test_issue_description_extracts_epic_plan_blocks(self):
+        """Epic plan descriptions survive the markdown to ADF to JiraIssue round trip."""
+        plan = """# Task Takeover Routing
+
+- Keep workflow identity in TaskTakeoverWorkflow.matches().
+- Preserve forge:managed and add a separate trigger label.
+
+```python
+def matches(issue):
+    return True
+```"""
+        adf = JiraClient._text_to_adf(plan)
+
+        issue = JiraIssue.from_api_response(
+            {
+                "key": "AISOS-1981",
+                "id": "10001",
+                "fields": {
+                    "summary": "Task Takeover Routing",
+                    "description": adf,
+                    "status": {"name": "To Do"},
+                    "issuetype": {"name": "Epic"},
+                },
+            }
+        )
+
+        assert "# Task Takeover Routing" in issue.description
+        assert "- Keep workflow identity" in issue.description
+        assert "- Preserve forge:managed" in issue.description
+        assert "```python" in issue.description
+        assert "def matches(issue):" in issue.description
+
+    def test_issue_description_preserves_unparsed_adf_node(self):
+        """Unknown ADF nodes should not disappear from issue descriptions."""
+        issue = JiraIssue.from_api_response(
+            {
+                "key": "AISOS-1981",
+                "id": "10001",
+                "fields": {
+                    "summary": "Unknown node",
+                    "description": {
+                        "type": "doc",
+                        "version": 1,
+                        "content": [
+                            {
+                                "type": "unsupportedWidget",
+                                "attrs": {"payload": "important raw context"},
+                            }
+                        ],
+                    },
+                    "status": {"name": "To Do"},
+                    "issuetype": {"name": "Epic"},
+                },
+            }
+        )
+
+        assert "unsupportedWidget" in issue.description
+        assert "important raw context" in issue.description
+
 
 @pytest.fixture
 def jira_client():