We propose to use the Short Wavelength Spectrometer to observe
 a sample of highly-obscured asymptotic giant branch stars in
 both the LMC and the SMC. These stars, which have been identified
 through near-IR follow-up observations of IRAS sources, are in
 the process of evolving from the AGB to become planetary nebulae.
 Although the classical limit to AGB evolution is M(bol)=-7.1,
 these giants have bolometric magnitudes as low as -6. The aim
 of our observations is to determine whether these stars have the
 11.6 micron SiC emission that characterises carbon-rich stars, or
 the broad 10 and 18 micron emission features of the O-rich M-type
 stars. This is of particular interest since, although carbon stars
 are common amongst the lower-luminosity AGB stars, very few are
 found brighter than M(bol)=-6, despite the theoretical expectation
 that the brighter, more massive stars should be more efficient at
 synthesising carbon. It is possible that envelope-burning may be
 responsible for maintaining C < O, but another possibility is that
 the higher opacity in C-rich envelopes may truncate AGB evolution
 through high mass-loss, thereby accounting for the general
 scarcity of luminous AGB stars. In the latter case, we would
 expect the highly-obscured, cocoon stars to have C-rich dust
 shells. Since all of our sample are optically invisible, ISO
 represents the only method of determining the chemical composition
 of the circumstellar dust-shells.