diff --git a/src/ccbot/transcript_format.py b/src/ccbot/transcript_format.py
index 8fbc7a19..61cc4a6e 100644
--- a/src/ccbot/transcript_format.py
+++ b/src/ccbot/transcript_format.py
@@ -135,10 +135,31 @@ def format_tool_use_summary(name: str, input_data: dict[str, Any] | Any) -> str:
     return f"**{name}**"
 
 
+# Claude Code wraps blocked / hook-rejected tool calls in a literal
+# ``<tool_use_error>…</tool_use_error>`` envelope inside the JSONL
+# tool_result content. The tags are an internal framing artefact —
+# rendering them verbatim ("Bash · Error: <tool_use_error>Blocked: …")
+# leaks implementation detail into the card head. Strip both halves
+# wherever they appear; the inner message is what the user wants.
+_TOOL_USE_ERROR_OPEN = "<tool_use_error>"
+_TOOL_USE_ERROR_CLOSE = "</tool_use_error>"
+
+
+def _strip_tool_use_error_envelope(text: str) -> str:
+    if _TOOL_USE_ERROR_OPEN not in text and _TOOL_USE_ERROR_CLOSE not in text:
+        return text
+    return text.replace(_TOOL_USE_ERROR_OPEN, "").replace(_TOOL_USE_ERROR_CLOSE, "")
+
+
 def extract_tool_result_text(content: list[Any] | Any) -> str:
-    """Concatenate all text fragments from a tool_result content block."""
+    """Concatenate all text fragments from a tool_result content block.
+
+    Strips the ``<tool_use_error>…</tool_use_error>`` envelope Claude Code
+    uses to wrap hook-rejected tool calls — those tags would otherwise
+    end up in the card head verbatim.
+    """
     if isinstance(content, str):
-        return content
+        return _strip_tool_use_error_envelope(content)
     if isinstance(content, list):
         parts: list[str] = []
         for item in content:
@@ -148,7 +169,7 @@ def extract_tool_result_text(content: list[Any] | Any) -> str:
                     parts.append(t)
             elif isinstance(item, str):
                 parts.append(item)
-        return "\n".join(parts)
+        return _strip_tool_use_error_envelope("\n".join(parts))
     return ""
 
 
diff --git a/tests/ccbot/test_transcript_format.py b/tests/ccbot/test_transcript_format.py
index 7598917e..0552d453 100644
--- a/tests/ccbot/test_transcript_format.py
+++ b/tests/ccbot/test_transcript_format.py
@@ -54,6 +54,38 @@ def test_skips_non_text_blocks(self) -> None:
         ]
         assert transcript_format.extract_tool_result_text(content) == "a\nb"
 
+    def test_strips_tool_use_error_envelope_string(self) -> None:
+        # Claude Code wraps hook-rejected results in
+        # ``<tool_use_error>...</tool_use_error>``. Without stripping,
+        # the literal tags ended up in card heads like
+        # "Bash · Error: <tool_use_error>Blocked: sleep 25 ...".
+        wrapped = (
+            "<tool_use_error>Blocked: sleep 25 followed by: cat /tmp/x. "
+            "To wait for a condition, use ...</tool_use_error>"
+        )
+        result = transcript_format.extract_tool_result_text(wrapped)
+        assert "<tool_use_error>" not in result
+        assert "</tool_use_error>" not in result
+        assert result.startswith("Blocked: sleep 25")
+
+    def test_strips_tool_use_error_envelope_in_list(self) -> None:
+        content = [
+            {
+                "type": "text",
+                "text": ("<tool_use_error>Blocked: rm -rf /</tool_use_error>"),
+            }
+        ]
+        result = transcript_format.extract_tool_result_text(content)
+        assert "<tool_use_error>" not in result
+        assert result == "Blocked: rm -rf /"
+
+    def test_no_envelope_is_passthrough(self) -> None:
+        # Make sure the strip path doesn't touch normal output.
+        assert (
+            transcript_format.extract_tool_result_text("normal output")
+            == "normal output"
+        )
+
 
 class TestExtractToolResultImages:
     def test_no_images_returns_none(self) -> None: