The central hypothesis of Relativity - both Special and General - is that at each point and time (at each space-time location), in each direction there is a finite non-zero invariant speed. Since this speed is associated with the speed of wave propagation in a vacuum - particularly of light - then it is usually referred to as "light speed".
Were you to depict this, in a space-time diagram, with the vertical direction being time-like and future-pointing, the locus of all trajectories that emanate from a given point and time at that speed would form an upwardly-directed cone having that space-time location as its vertex. That's its Future Light Cone. Similarly, the locus of all trajectories that arrive at that space-time location at that speed would form a downward-directed cone - the Past Light Cone.
Correspondingly, the geometry is enmeshed with a field of light cones, one pair for each place and time.
The field of light cones - by itself - is actually sufficient to specify the metric for the geometry ... up to a conformal transformation. So, it embodies the conformal part of the metric (i.e the class of metrics that is conformally equivalent to the given metric).
The metrics that correspond to this field will have a set of components that may be arrayed as a matrix. At each space-time location, the matrix for that metric may be reduced to a diagonal form either as (+1, +1, +1, -1), if the metric is treated as a metric for proper distances, or as (+1, -1, -1, -1), if it is treated as a metric for proper times.
Such a metric is said to have "Lorentzian signature" and a space-time which has - at every one of its locations - a Lorentzian signature is a Lorentzian manifold and is precisely what we actually mean by the term "space-time". For other dimensional geometries, the term is normally understood to apply to metrics whose signature is (+,-,-,...,-) or (+,+,...,+,-), where all but one of the dimensions is "space-like" and the other is "time-like".
It's worth pointing out that Hawking and Hartle dealt with signature-changing geometries, where the Lorentzian part of the geometry is connected to a locally 4D-Euclidean part, where the signature is (+,+,+,+). There are other people who work with signature-changing geometries, such as Mansouri. Technically, this violates the central axiom of Relativity, because now you only have a sub-space that is Lorentzian. The interface where it connects up with the non-Lorentzian part would be the boundary of the zone where, strictly speaking, the applicability of Relativity is confined to.
If the light cones can be mapped congruently onto themselves by spatial translations, then they are spatially homogeneous. If they can be mapped congruently onto themselves under time translations, then they are homogeneous in time. In that sense, you could say that the speed of light would then be constant throughout space and in time. If they can be mapped congruently onto themselves by reorientation of the spatial axes, then they are isotropic and you could say (in that sense) that light speed is direction-independent. Finally, if the structure of the light cones remains invariant under a set of transformations that uniformly change the moving state of an observer (these transforms are generically called "boosts", with the Lorentz transform and Galilei transform being two cases in point), then one could say that light speed is observer-independent. The boosts would all be Lorentz transforms.
A field of light cones that has all four set of symmetries is specified up to a conformal transformation by the Minkowski metric. That distinguishes Special Relativity from General Relativity.
So, both Special and General Relativity stipulate that the underlying geometry be Lorentzian ... or at the very least that a subspace of it that contains our world does. Special Relativity goes one step further in asserting that the field of light cones is constant and symmetric so that we may also assert (in this sense) that light speed is a constant. A set of global transformations exist that map the light-speed trajectories at any given space-time location to those of another, and which remain invariant under an arbitrary change in orientation and an arbitrary boost.
If you count the symmetries: 3 degrees of spatial translation, 1 degree of time translation, 3 set of axes for rotation, and 3 sets of directions for boosts, that's 10 in all. Since the underlying metric is Lorentzian, then the transformations together gives you a realization of the Poincaré group. So, the metric in Special Relativity has the Poincaré group as a transformation group, as do the corresponding field of light cones.
The Poincaré group, with conformal transforms added in, forms the Conformal Group, which contains 5 additional sets of transforms. I'm a little fuzzy on what the relation of the light cone field to this group is, though I'm pretty sure it has the full Conformal Group as its transform group, given the previous discussion.