The problem with using cell sense lines for this is that often, common BMSs can't detect that a cell is disconnected, or at 0v; they only detect that it's crossed below the LVC point or above the HVC point.
Some of them actually recognize that cells are disconnected, so using a thermal switch between one cell and the BMS, would disconnect that cell when the switch opens, which would trigger the BMS to shutoff output or input.
Some of them don't know a cell is disconnected, but still recognize a cell is dead at 0v.
In the first case, you'd have to build electronics to cause it to indicate to the BMS that the cell was below LVC, but not so far below that it just ignores the problem, *and* that would not actually affect the cell itself.
In the second case, it would jsut work.
In the third case, it would work if you used it to short the cell balance wire to ground, but you'd need electronics to isolate the cell first.
It would be simpler to locate the gate drive circuit for the input and output FETs, and put thermal sensors in series with *those*, so the BMS can't turn those FETs on. (and use a high resistance between the gate and ground so when the switch opens no RF noise / etc can accidentaly turn the FETs on).
