re_behaviour: the emulator check for given functions of a worktree

harbor/mcp_server.py

@@ -539,6 +539,20 @@ def re_match_program(targets: str = "", decomp: str | None = None) -> dict:
         raise ToolError(str(error)) from None
 
 
+@mcp.tool()
+def re_behaviour(targets: str, decomp: str | None = None) -> list:
+    """The emulator check: run the game's own machine code and your compiled C of each function (up to 30
+    addresses; decomp = your worktree) from the same registers, stack and memory, 12 trials each, and compare the
+    result, every memory write, the calls made and their arguments. verdict 'same' means it behaves like the game;
+    'differs' gives the first divergence in words (e.g. 'returns 1 vs 0', 'original writes 0x86BAC0=5, ours 4');
+    'weak' means no trial ran to the end in both. Use it before submitting anything that does not byte-match."""
+    try:
+        from .reverse import behaviour_check
+        return behaviour_check(harbor().project.reverse, targets, decomp)
+    except (HarborError, projects.ProjectError) as error:
+        raise ToolError(str(error)) from None
+
+
 @mcp.tool()
 def re_similar(target: str, k: int = 10, min_score: float = 0.7) -> dict:
     """Functions whose instructions look like this one (normalized: addresses blanked, registers and small