It sounds like you may be dancing around the possibility that there are in fact two types of sequences: discrete and continuous. Each one has their own unique representation and behavior model and they are actually quite complimentary to one another.
A discrete sequence is used to represent transitions between clearly discernible states (C0 discontinuous). These sequences are stored as discrete transitions between the SDRs for the states. Conceptually, this can be considered a form of graph representation, and is probably the most efficient way to model sequences with a finite number of transitions. This is the classical HTM sequence model, and I believe it is the appropriate model to invoke when considering high-level planning.
A continuous sequence is used to represent any feature or space that could be considered continuous (C0) and/or differentiable (C1) w.r.t. some parameter (spatial, temporal, or conceptual). This type of feature space is probably best represented by coupling grid cell modules to one of the discrete feature SDRs described above. This coupling allows for interpolation in and around a collection of discrete features in an otherwise continuous state space. I can think of multiple ways to implement this in TBT; the most obvious one being a quasi-stable SDR (in L2/3) that represents a canonical feature (L4) at a pose (L6) coupled to a set of grid-cell modules that represents some small offset from this canonical representation. This SDR could then conceivably transition seamlessly to another adjacent SDR and its continuous space (i.e. reanchoring).
These two representations are complimentary to one another. When planning our movements (or thinking), we first operate at a high-level in the discrete sense. Here we are attempting to identify the landmark states that will take us from some initial state to some final state. Path planning is literally the process of identifying a set of intermediate states that must be passed through. In other words, identify what intermediate states are accessible from the initial state and potentially getting closer to the final state. Because this is a graph model rather than a continuous space model, we are free to have state transitions at multiple levels (ref. Tolman-Eichenbaum Machine).
As part of this process, at any time, we can mentally anchor ourself to anyone of these intermediate states and consider a more refined discrete sequence planning (e.g. sub-stepping) or possibly explore the continuous space around that state (e.g. gradient sampling). This allows us to consider additional factors that might impact the broader path. From this we can not only construct a nearly optimal path, but also consider reasonable alternatives. (e.g. What route will I take home from work tonight if my regular route is congested due to the parade downtown?)
Once we have sufficiently considered and settled on a plan (or at least a preliminary plan), the go/no-go decision is reached and we start to navigate through the plan. Starting with the initial state we instantiate the continuous representation around that state and begin to move in the direction of the first way-point (landmark state). Once we get close enough to the next way-point, we reanchor our current state to it and continue to navigate through the continuous space of this new state. If at any point, our perception of the state space is not aligned with our imagined predictions during planning, we can immediately reevaluate the plan considering the newly acquired information.
So, for the stapler example. You may have several canonical SDRs to which you can anchor your perception (open, half-opened, closed, compressed). Around each of these, there is a continuous morphological space that you have experienced in the past and for which you have a good perceptual/behavioral model. When planning to close the stapler, you can easily imagine going from the open state to the closed state, but you can also imagine stopping at any intermediate state along the way. However, you are always aware that there is a continuous set of states that must be passed through to get from the open to close state. While I can imagine the stapler instantly going from open to closed, I also understand that the probability of that actually happening is vanishingly small. (i.e. I would be very surprised if I observed it behaving that way.)
