fix(security): gate bare and mcp-qualified email names together; stop executing Markdown info strings

Review follow-up on #3681 (thanks @RaresKeY):

1. P1: execute_tool_block() checked disabled_tools / the turn ToolPolicy
   only against the incoming block name, then the bare-email branch
   qualified it to mcp__email__<name> and called the MCP manager. Plan
   mode and the MCP settings toggle write the QUALIFIED name into the
   denylist, so a bare fence like ```list_emails``` sailed past a
   mcp__email__list_emails entry. Both gates now match on both
   spellings (bare <-> mcp__email__-qualified), in either direction.

2. P2: the relaxed fence regex accepted arbitrary same-line text after
   a recognized tag, which made ordinary Markdown info strings
   executable: ```python title="example.py" ran as a python tool call.
   Same-line content now only counts as tool input when it starts with
   { or [ (JSON args); anything else leaves the fence as display text,
   and strip_tool_blocks mirrors that (the fence stays visible).

Tests: disabled-tools alias regression (qualified entry blocks bare
name and vice versa, never reaching the MCP manager), ToolPolicy alias
regression, python/bash title="..." non-execution + display retention,
and inline JSON-array args still parsing.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

src/tool_execution.py

@@ -559,6 +559,19 @@ async def _execute_tool_block_impl(
     tool = block.tool_type
     content = block.content
 
+    # A bare email tool name is an alias for its MCP-qualified form (the
+    # dispatch below routes it to mcp__email__<name>). Policy sources spell
+    # email tools either way — plan mode and the MCP settings toggle write the
+    # qualified name into the denylist, chat-level toggles the bare one — so
+    # the block/disable gates below must match on BOTH spellings. Gating only
+    # the name the model happened to emit lets a bare fence slip past a
+    # qualified denylist entry (and vice versa).
+    policy_names = {tool}
+    if tool in BUILTIN_EMAIL_TOOLS:
+        policy_names.add(f"mcp__email__{tool}")
+    elif tool.startswith("mcp__email__") and tool[len("mcp__email__"):] in BUILTIN_EMAIL_TOOLS:
+        policy_names.add(tool[len("mcp__email__"):])
+
     # Misformatted tool call detection: model put JSON inside ```python``` (or
     # similar) without naming the tool. Common with MiniMax-style outputs.
     # Return a helpful error so the model retries with the correct format.
@@ -586,13 +599,13 @@ async def _execute_tool_block_impl(
             pass
 
     # Reject tools that the user has disabled for this request
-    if disabled_tools and tool in disabled_tools:
+    if disabled_tools and not policy_names.isdisjoint(disabled_tools):
         desc = f"{tool}: BLOCKED"
         result = {"error": f"Tool '{tool}' is disabled by user.", "exit_code": 1}
         logger.info(f"Tool blocked by user: {tool}")
         return desc, result
 
-    if tool_policy and tool_policy.blocks(tool):
+    if tool_policy and any(tool_policy.blocks(name) for name in policy_names):
         desc = f"{tool}: BLOCKED"
         result = {
             "error": f"Execution of tool '{tool}' is forbade by the active guide-only policy.",

src/tool_parsing.py

@@ -19,13 +19,17 @@ logger = logging.getLogger(__name__)
 # Regex patterns
 # ---------------------------------------------------------------------------
 
-# Pattern 1: ```bash ... ``` fenced code blocks. The tag may be followed by
-# inline args on the same line (```list_email_accounts {}) or a newline.
+# Pattern 1: ```bash ... ``` fenced code blocks. The tag may be followed by a
+# newline (classic form) or by inline JSON args on the same line
+# (```list_email_accounts {}). Same-line content is accepted ONLY when it
+# starts with { or [ — anything else after the tag is a Markdown info string
+# (```python title="example.py"), which must stay display text rather than
+# become executable tool input.
 # (?![\w-]) keeps the alternation from prefix-matching longer fence tags:
 # without it, ```python3 would match as tool "python" with content "3\n..."
 # and execute as code.
 _TOOL_BLOCK_RE = re.compile(
-    r"```(" + "|".join(TOOL_TAGS) + r")(?![\w-])[ \t]*\n?([\s\S]*?)```",
+    r"```(" + "|".join(TOOL_TAGS) + r")(?![\w-])[ \t]*(?=\r?\n|[{\[])\r?\n?([\s\S]*?)```",
     re.IGNORECASE,
 )