Are Marasuchus, SMNS 12352 and Trialestes theropods?

No, of course not, as we saw in my earlier post where correcting codings and adding data excludes them from the clade.  But Peters thinks they belong.  Let's go through his supporting characters below.

Figure 1. Trialestes holotype skull (left) modified after Reig (1963), and forelimb (right) after Bonaparte (1978).  Unknown portions of skull and hypothetical dotted sutures drawn by Reig removed, except around the external mandibular fenestra which is apparently present but of unknown shape.  Note these are the only illustrations available, and no photos or further information on the skull not obvious from Reig's drawing, so when Peters codes things like skull width or palatal characters, it's just his imagination.

Theropoda (Trialestes, Herrerasaurus, Marasuchus, SMNS 12352, Tawa, Coelophysis)
3. Skull less than 1.2 times wider than tall.  This was miscoded as present in Trialestes and Marasuchus (see figures 1 and 2), when they cannot actually be coded.  Also, this is present in all examined dinosaurs except Lesothosaurus, so many 'phytodinosaurs' were miscoded.
13. Snout convex and angled over external naris.  This was miscoded as present in Trialestes and Marasuchus, when they cannot actually be coded (see figures 1 and 2).  One side of Herrerasaurus' snout is smoothly curved instead of angled, so it can't be coded unambiguously.  As more derived theropods reverse this in Peters' tree, this completely eliminates the character's validity.
55. Posterior jugal process descends.  This was miscoded as missing in sauropodomorphs, Silesaurus and Lotosaurus.  In actuality, all examined dinosauriforms have it except Daemonosaurus, Heterodontosaurus and some Coelophysis individuals.
63. Jugal posterior process extends more than halfway down cheek.  This was miscoded as absent in Gracilisuchus, which would make it a symplesiomorphy lost in phytodinosaurs instead.
89. Maxilla palatal process present.  This is actually present in all examined taxa except seemingly Turfanosuchus, but Peters miscoded sauropodomorphs, ornithischians and Gracilisuchus as lacking it.  Amusingly, two of the three 'phytodinosaurs' he coded (Pisanosaurus and Pantydraco) actually can not be, and neither can Trialestes (see figure 1).
92. Choanae "medial rostral pinch", or as Peters has explained via pers. comm., choanae with posterior end angled medially due to the palatine lining some of the lateral margin, and maxillae anteriorly pinched instead of parallel.  You can see this is a composite character, and indeed the choanae are lined by the palatines laterally in their posterior part in all taxa examined here.  Ironically, Coelophysis and Herrerasaurus actually have more parallel maxillae than the other two taxa Peters codes that are actually scorable- Lesothosaurus and Silesaurus.  Trialestes cannot be coded from published data (see figure 1), while SMNS 12352 has more pinched maxillae.  Once all of the taxa Peters didn't code or include are scored, subparallel maxillae are present in Arizonasaurus and Dinosauria, except for ornithischians besides Heterodontosaurus (which works with heterodontosaurids being the basalmost ornithischians).
206. Metatarsus compact, which as Peters codes it seems to involve metatarsals II-IV contacting over most of their shafts.  Miscoded as absent in Silesaurus and Pisanosaurus, and given the wrong polarity in Ornithischia because he has Scelidosaurus most basal (when Pisanosaurus, heterodontosaurids and Lesothosaurus are most basal, a compact metatarsus is primitive for the clade).  This is actually primitive for dinosauriforms and lost in sauropodomorphs and thyreophorans.  Also, Trialestes was coded for it based on a dinosaurian foot.  Ironically, that metatarsus is not restored as being compact anyway.
216. Pedal phalanges of digit IV shorter than metatarsal IV.  Again, Trialestes was coded based on the wrongly referred foot.  Even that foot only preserves phalanx IV-4, part of IV-3 and part of ungual IV, and metatarsal IV is broken so its length is unknown.  So it couldn't be coded anyway.  Herrerasaurus was not coded but lacks the character, so it would actually support the Marasuchus+Coelophysis node in Peters' phylogeny.  But it's lacking in the next most derived scorable taxon, Dilophosaurus.  As Tawa cannot be coded yet, this makes it ambiguous.  It was also miscoded as absent in Saturnalia, though Pantydraco was left uncoded and lacks it.
Results- All of these only optimize to Theropoda due to miscodings.  Trialestes only has two, which are both primitive once taxa are coded correctly.  Marasuchus also only has two, one of which is primitive once taxa are correctly coded, though the other which would be valid for joining it with Coelophysis.  SMNS 12352 has two, but one is present in almost all taxa and the other is lacking in the actual theropods.

(Herrerasaurus, Marasuchus, SMNS 12352, Tawa, Coelophysis)
This is Theropoda in Peters' more recent tree.
161. Scapulocoracoid fused.  Miscoded in Tawa, while Coelophysis can have unfused bones too.  So if coded correctly, would only ambiguously support this clade.  Also miscoded as absent in Saturnalia.  Honestly, this near certainly varies with age, so related taxa like Pampadromeus, Panphagia and Pantydraco with unfused elements may just not be fully grown.
179. Manual digit IV reduced in width.  In Peters' tree, this is convergent with sauropodomorphs+ornithischians.  But that's only because he miscodes it as absent in Pisanosaurus (doesn't even preserve digit IV), Effigia and Poposaurus.  If coded correctly in his tree, it would diagnose Dinosauria and reverse in Lotosaurus.  Also note SMNS 12352 lacks it (see figure 3), though Peters doesn't code its hand.
180. Manual phalanx II-2 longer than II-1.  Peters again doesn't code SMNS 12352's manus, but it lacks the character.  While this is actually usually thought to be a theropod character, its presence in taxa Peters doesn't include like Guaibasaurus, Efraasia and Eocursor might suggest it's primitive to Dinosauria instead.
182. Metacarpal V absent.  This is miscoded, as both Herrerasaurus and Coelophysis have metacarpal V.  So does SMNS 13252 (see figure 3), which was again not coded for manual characters.
191. At least semiperforated acetabulum.  This is another character that is convergent in his tree with sauropodomorphs+ornithischians.  But that's only because he miscodes poposaurs, Pampadromaeus and Panphagia as lacking it.  It's actually present in all dinosaurs (and convergently poposaurs), while Marasuchus has at most a tiny slit that Silesaurus might also have (note the non-included Sacisaurus has it, but Asilisaurus does not, so the exact sequence of acquisition at the base of Dinosauriformes is uncertain).
194. Large pubic boot.  This was miscoded as present in Marasuchus, when the pubic apron of that taxon is actually only folded over to make it appear as if the bone expands distally.  As Coelophysis lacks it, this makes it shared between only Herrerasaurus and Tawa.  I have no issue with them being related, though its absence in the unincluded Eodromaeus complicates things.
Results- Marasuchus only has two, one of which is primitive for Dinosauria (at least) once it is coded correctly, and the other which is an ambiguous synapomorphy once coded correctly and may be ontogenetic.  It also lacks one.  SMNS 12352 has none and lacks two, though it was not even coded for them.

Figure 2. Skull material of Marasuchus specimen PVL 3870, modified after Bonaparte (1975). Maxilla in medial view, though Bonaparte thought the anterior fragment was the premaxilla and the posterior margin was the antorbital fenestra, this isn't true.  The postorbital was misidentified as a quadrate by Bonaparte and has been moved here.  These new interpretations are all from Sereno and Arcucci (1994).

(Marasuchus, SMNS 12352, Tawa, Coelophysis)
11. Nasals with parallel sides in dorsal view.  Marasuchus doesn't even preserve nasals so is miscoded (see figure 2), and their shape hasn't been reported in Tawa, so that's miscoded too.  Furthermore, SMNS 12352 and Coelophysis actually have posteriorly expanding nasals like Herrerasaurus, so every taxon was miscoded.  And lest you think posteriorly expanding nasals could be a theropod character of SMNS 12352, note the crocodylomorph Terrestrisuchus also has them.
146. Second sacral rib not bifurcated.  This is here because Peters miscodes all other scored taxa as having a bifurcate rib, but no archosaur actually has one.  Confusing.  Maybe he misunderstood the character?
147. Second caudal transverse process longer than centrum.  This is miscoded as present in the two scored taxa- Marasuchus and Coelophysis.  It's actually only present in Hexinlusaurus among included taxa, though Peters didn't score that.
152. Proximal caudal neural spines shorter than centrum height.  This was miscoded as being present in Coelophysis.  Thus it would only ambiguously support this clade.  Even then, the coding in Tawa is based on a schematic skeletal reconstruction, while the caudals of Marasuchus are preserved to show the ventrolateral faces and artificially shorten the spines due to perspective.
154. Mid caudal centra (here defined as vertebrae which reduce the transverse processes and spines but not to mere ridges) at least three times longer than tall.  *Gasp*  For once, a valid character.  Marasuchus does indeed share this with Tawa and avepods, as well as the non-included Chindesaurus and Eodromaeus.
208. Metatarsal I less than 50% of metatarsal III.  This is miscoded and is absent in Marasuchus.  Among included taxa, it's only present in Coelophysis and some Lesothosaurus individuals.
Results- Marasuchus has two, one of which is valid and the other which is present in all archosaurs.  It may have another, but that would only connect it with Tawa if THAT taxon's reconstruction is correct, not Coelophysis.  It also lacks two.  SMNS 12352 lacks one and doesn't have any.

Figure 3. SMNS 12352 snout in lateral view (C) and manus in extensor view (D), after Sereno and Wild (1992).

(SMNS 12352, Tawa, Coelophysis)
13. Snout triangular due to having only a slight convexity over the naris.  SMNS 12352 doesn't preserve the snout tip, but the low angled dorsal surface suggests it had a notable convexity if anything, so is miscoded.  This may be a useful synapomorphy for Tawa and Coelophysis, and the miscoded Daemonosaurus.
20. External naris angled 30 degrees or less from horizontal.  SMNS 12352 doesn't preserve any part of the external naris, so was miscoded (see figure 3).  This again actually does diagnose Tawa and Coelophysis, and is also found in the non-included Eoraptor.
23. External naris placed at snout tip.  This is contrasted with adjacent taxa being coded as having a naris "at snout tip but dorsal", which highlights its composite nature coding for both anteroposterior and dorsoventral placement.  But none of that matters here, as SMNS 12352 doesn't preserve any part of the external naris, so can't be coded regardless (see figure 3).
Results- SMNS 12352 can't be coded for two and probably lacks the third. 

Total Theropod Results- Of eight characters, Trialestes only has two, and both are primitive once miscodings are corrected.  Small wonder it's no longer a theropod in Peters' newest tree.  I would say the wrongly referred foot influenced it, but even that lacks Peters' theropod characters.
Of twenty characters, Marasuchus has five, might have two more, and lacks three.  Of the first seven, three are primitive for dinosaurs once coded correctly in other taxa, and three more are absent in a basal theropod so are ambiguous (one is also probably ontogenetic).  The one valid character is that it has elongate mid caudals, but as it lacks e.g. more than ten dinosauriform and saurischian characters neglected by Peters, this is hardly meaningful. Edit: In fact, it takes 11 more steps to make Marasuchus a theropod in my corrected Peters matrix with added characters and taxa, and 34 more steps in Nesbitt's (2011) larger matrix.
Of twenty-three characters, SMNS 12352 has two.  Maxillary palatal processes are present in all archosaurs, and a pinched palate is lacking in the actual theropods.  It lacks five, and while one of those is actually lacking in theropods as well, Terrestrisuchus has the same state.  I should note that Terrestrisuchus itself falls within Theropoda if only skull characters are coded, so Peters' characters cannot distinguish theropod and sphenosuchian skulls even when correctly coded.  This makes it even more problematic that Peters didn't code SMNS 12352's manus. Edit: It takes 5 more steps to make SMNS 12352 a theropod in my corrected Peters matrix with added characters and taxa.

As you can see, Peters' characters provide no evidence Trialestes or SMNS 12352 are theropodan, and only a single character supporting Marasuchus as one, and that's vastly outweighed by the dinosaurian and saurischian characters he doesn't include that have been standard in the literature for two decades.