Component: finbot/mcp/servers/systemutils/server.py → run_diagnostics
Root cause:

def run_diagnostics(command: str) -> dict[str, Any]:
    # no allowlist check — any string accepted
    return {"command": command, "status": "completed", ...}

Steps to reproduce:

1. Call run_diagnostics(command='disk_usage; rm -rf /data')

Expected: error — shell injection characters rejected
Actual: command recorded as completed with exit_code=0

How to execute:

pytest tests/unit/mcp/test_systemutils.py::TestRunDiagnostics::test_su_diag_003_shell_injection_accepted_without_validation -v

Proposed fix: Enforce a command allowlist at the top of run_diagnostics:

ALLOWED_COMMANDS = {"disk_usage", "memory_check", "network_status", "process_list"}
if command not in ALLOWED_COMMANDS:
    raise ValueError(f"Command '{command}' is not in the list of permitted diagnostic commands")

Impact: No code-level guard if the mock is replaced with real execution. CTF detectors flag shell injection attempts as dangerous operations.

Acceptance criteria:

• test_su_diag_003_shell_injection_accepted_without_validation passes (exception raised for injection command)
• test_su_diag_001_returns_expected_fields continues to pass (valid commands still work)