diff --git a/docker/clawscan-runtime/Dockerfile b/docker/clawscan-runtime/Dockerfile
index baeedcb..1bb4a92 100644
--- a/docker/clawscan-runtime/Dockerfile
+++ b/docker/clawscan-runtime/Dockerfile
@@ -41,7 +41,7 @@ RUN git init /tmp/skillspector \
 RUN python -m pip install --no-cache-dir --upgrade pip uv \
     && python -m pip install --no-cache-dir /tmp/skillspector pytest \
       "cisco-ai-skill-scanner==$CISCO_AI_SKILL_SCANNER_VERSION" \
-    && python -m pytest -q /tmp/clawscan-adapter-tests/test_skillspector_responses.py /tmp/clawscan-adapter-tests/test_skillspector_applicability.py /tmp/clawscan-adapter-tests/test_skillspector_accounting.py /tmp/clawscan-adapter-tests/test_skillspector_frontmatter_json.py /tmp/clawscan-adapter-tests/test_skillspector_osv_evidence.py /tmp/clawscan-adapter-tests/test_skillspector_pdf.py /tmp/clawscan-adapter-tests/test_skillspector_markdown_containers.py /tmp/clawscan-adapter-tests/test_skillspector_semantic_packets.py /tmp/clawscan-adapter-tests/test_skillspector_meta_packets.py /tmp/clawscan-adapter-tests/test_skillspector_security_view_cache.py /tmp/clawscan-adapter-tests/test_skillspector_ledger_capacity.py /tmp/clawscan-adapter-tests/test_skillspector_table_markers.py \
+    && python -m pytest -q /tmp/clawscan-adapter-tests/test_skillspector_responses.py /tmp/clawscan-adapter-tests/test_skillspector_applicability.py /tmp/clawscan-adapter-tests/test_skillspector_accounting.py /tmp/clawscan-adapter-tests/test_skillspector_frontmatter_json.py /tmp/clawscan-adapter-tests/test_skillspector_osv_evidence.py /tmp/clawscan-adapter-tests/test_skillspector_pdf.py /tmp/clawscan-adapter-tests/test_skillspector_markdown_containers.py /tmp/clawscan-adapter-tests/test_skillspector_semantic_packets.py /tmp/clawscan-adapter-tests/test_skillspector_meta_packets.py /tmp/clawscan-adapter-tests/test_skillspector_security_view_cache.py /tmp/clawscan-adapter-tests/test_skillspector_ledger_capacity.py /tmp/clawscan-adapter-tests/test_skillspector_table_markers.py /tmp/clawscan-adapter-tests/test_skillspector_fence_boundaries.py \
     && uv tool install /tmp/aig/skill-scan --with pytest \
     && /root/.local/share/uv/tools/aig-skill-scan/bin/python -m pytest -q /tmp/clawscan-adapter-tests/test_aig_responses.py \
     && uv tool install "snyk-agent-scan==$SNYK_AGENT_SCAN_VERSION" \
diff --git a/docker/clawscan-runtime/patches/skillspector-responses.patch b/docker/clawscan-runtime/patches/skillspector-responses.patch
index bb50878..bda04c2 100644
--- a/docker/clawscan-runtime/patches/skillspector-responses.patch
+++ b/docker/clawscan-runtime/patches/skillspector-responses.patch
@@ -1911,7 +1911,7 @@ index 74c304a..b8bc498 100644
 +    assert len(requests) == 2
 +    assert len(osv._cache) == 1
 diff --git a/src/skillspector/nodes/analyzers/static_patterns_tool_misuse.py b/src/skillspector/nodes/analyzers/static_patterns_tool_misuse.py
-index c323c84..5db9eca 100644
+index c323c84..49e4a5f 100644
 --- a/src/skillspector/nodes/analyzers/static_patterns_tool_misuse.py
 +++ b/src/skillspector/nodes/analyzers/static_patterns_tool_misuse.py
 @@ -41,6 +41,7 @@ from skillspector.artifacts import (
@@ -1922,15 +1922,30 @@ index c323c84..5db9eca 100644
  from skillspector.logging_config import get_logger
  from skillspector.models import AnalyzerFinding, Finding, Location, Severity
  from skillspector.python_ast import ParsedPythonFile, parse_python_source
-@@ -4847,6 +4848,7 @@ def _markdown_shell_text(
+@@ -4847,6 +4848,8 @@ def _markdown_shell_text(
      table_columns: int | None = None
      table_minimum_indent = 0
      table_delimiter_line = -1
 +    table_ranges: list[tuple[int, int]] = []
++    fence_ranges: list[tuple[int, int]] = []
      offset = 0
      lines = content.splitlines(keepends=True)
      for line_index, line in enumerate(lines):
-@@ -4997,6 +4999,7 @@ def _markdown_shell_text(
+@@ -4921,6 +4924,13 @@ def _markdown_shell_text(
+             if (html_end and html_end in leading.lower()) or (not html_end and not leading):
+                 html_end = None
+         elif fence is not None:
++            # A canonical container parse may end a list fence at dedentation.
++            # Keep this projection's stricter literal boundary when refining
++            # inline delimiters, just as for the table boundaries below.
++            if fence_ranges and fence_ranges[-1][1] == offset:
++                fence_ranges[-1] = (fence_ranges[-1][0], offset + len(line))
++            else:
++                fence_ranges.append((offset, offset + len(line)))
+             fence_body = stripped.lstrip(" \t")
+             closing_fence = MARKDOWN_FENCE_CLOSE.fullmatch(fence_body)
+             if (
+@@ -4997,6 +5007,7 @@ def _markdown_shell_text(
                          table_minimum_indent = minimum_indent
                          table_delimiter_line = line_index + 1
                  if table_columns is not None:
@@ -1938,7 +1953,7 @@ index c323c84..5db9eca 100644
                      mask_inline_delimiters()
                      if line_index != table_delimiter_line:
                          cells = _markdown_table_cells(stripped, prefix, check_runtime)
-@@ -5069,6 +5072,13 @@ def _markdown_shell_text(
+@@ -5069,6 +5080,13 @@ def _markdown_shell_text(
              paragraph_list_indent = 0
          offset += len(line)
      mask_inline_delimiters()
@@ -1946,13 +1961,13 @@ index c323c84..5db9eca 100644
 +        from skillspector.markdown_tokens import MarkdownCodeSpanClosers
 +
 +        for delimiter in MarkdownCodeSpanClosers(
-+            content, check_runtime, excluded_ranges=tuple(table_ranges)
++            content, check_runtime, excluded_ranges=tuple(table_ranges + fence_ranges)
 +        ).delimiter_offsets():
 +            output[delimiter] = " "
      projected = "".join(output)
      if prose_apostrophes:
          # A prose-looking line can still be inside an earlier shell word, such
-@@ -5152,6 +5162,7 @@ def _has_bounded_parse_exhaustion(
+@@ -5152,6 +5170,7 @@ def _has_bounded_parse_exhaustion(
      if file_type == "markdown":
          if complete_context:
              json_strings = validated_json_string_spans(content, check_runtime)
diff --git a/docker/clawscan-runtime/tests/test_skillspector_fence_boundaries.py b/docker/clawscan-runtime/tests/test_skillspector_fence_boundaries.py
new file mode 100644
index 0000000..6d418ea
--- /dev/null
+++ b/docker/clawscan-runtime/tests/test_skillspector_fence_boundaries.py
@@ -0,0 +1,30 @@
+"""Conservative shell fences keep their delimiters and bounded exclusions."""
+
+import pytest
+
+from skillspector.markdown_tokens import MarkdownCodeSpanClosers
+from skillspector.nodes.analyzers import static_patterns_tool_misuse as tool_misuse
+
+
+@pytest.mark.parametrize("opening,indent", [("   ```sh", "      "), ("- - ```sh", "        ")])
+def test_deeply_indented_marker_does_not_close_fence(opening, indent):
+    command = "`$(resolve_tool).example` -rf /"
+    content = opening + "\n" + indent + "```\n" + command + "\n```\n"
+    projected = tool_misuse._markdown_shell_text(content, lambda: None)
+    assert command in projected
+
+
+def test_fence_ownership_exclusions_are_bounded_by_blocks(monkeypatch):
+    content = "- ```sh\n" + "\n" * 10_000 + "  ```\n\n"
+    content += "Use `$(hostname).example` for the host.\n\n" * 100
+    exclusions = []
+    original = MarkdownCodeSpanClosers.__init__
+
+    def observe(self, *args, **kwargs):
+        exclusions.append(len(kwargs.get("excluded_ranges", ())))
+        original(self, *args, **kwargs)
+
+    monkeypatch.setattr(MarkdownCodeSpanClosers, "__init__", observe)
+    projected = tool_misuse._markdown_shell_text(content, lambda: None)
+    assert exclusions and max(exclusions) <= 1
+    assert "$(hostname).example" in projected