Non-Combat Skills / Dialog
You need to decide how you plan to handle non-combat abilities. The Mass Effect games seem to remove more and more in terms of non-combat abilities that require character stats/abilities. By ME3, all that was left was a dialog “skill” that generally only affected the final outcome of a situation after it was already physically (gameplay-wise) resolved. You couldn’t talk your way into an inaccessible room, you couldn’t investigate and find a way around a room full of guards. By focusing solely on gun battles and one-on-one dialog to the exclusion of pretty much all else you lose a lot of variety in terms of the kinds of scenarios you can put the player in. At the same time, if you put non-combat abilities back in you can have more freedom to play around with more unique/difficult combat situations. Maybe in one level there is a vault that requires really high electronics to open, but then in another level you can put an incredibly difficult optional “mini-boss” that you have to fight solo so that people who specced combat can have their day.
Another option would be to make traditionally “non-combat” skills usable in combat. Maybe you can hack a door or a lift to get an advantageous position for a really hard fight but you have to do it in real-time while under fire. In a situation like this maybe higher electronics reduces the number of steps you have to do, or provides a slight time-dilation effect. This way a character can go for a “brains over brawn” style that isn’t just talking the opponent down through dialog.
I am against the paragon/renegade system in general. It forces situations that should be ambiguous at best into “nice” and “mean” choices, which greatly diminishes the impact of these choices. It also shapes the creation of the situations themselves: decisions have to be written to have a paragon and renegade solution rather than solutions than would make the most logical sense in the situation. This makes all of the big decisions tend towards extreme either/or choices: kill/don’t kill, help/don’t help, save this or that. Choices between two different goods or two different evils would be very interesting, but paragon/renegade really prevents these types of choices; it stops being a difficult moral choice when one of the two evils is “paragon” or one of the two goods is “renegade.”
Dialog shouldn’t always have both paragon and renegade options for the same choice (in fact I believe the less often this happens the better); even more so when both options have the exact same outcome (i.e. not having to fight somebody).
Less use of paragon/renegade as the “third option” that is quite obviously the “best” choice (especially for “hard” choices). The use of “third option” is a little more acceptable if you don’t have both a paragon and renegade “third” option for the same choice (or at least very rarely), so being either 100% paragon or 100% renegade doesn’t give you an automatic “out” of any hard choice. Even better would be if the paragon/renegade choices aren’t always “good” for the given situation (or even at all). Maybe a cruel adversary laughs at your renegade threats of violence and kills the hostages, maybe your non-violent paragon attempts at interrogation fail to get the information you need in time and the bomb goes off. Anything that forces the player to actually consider their words/actions rather than just immediately go for the “best” red/blue response would be a step in the right direction.
A side note: there need to be more immediate negative consequences for choices in general. Paragons let a killer go and they (surprise) kill people, but it is always off-screen and never in a way that actually affected the game. Renegades shoot first and ask questions later, but it never turns out that they guy they threw out the window had the keycard they needed.
You could have the order you talk to people/access information matter in terms of the dialog you get. Maybe somebody tries to lie to you but you had previously found a recording that shows him doing it. You actually did this pretty well in ME2 with the blue sun mercenary in samara’s recruitment mission: if you talked to everybody you found out that they have to kill somebody to get their armor, so you know when she claims to be innocent she is lying. It would have been even better if you got an additional dialog option to confront her based on having acquired that knowledge beforehand though.
Combat (General)
Don’t always have the enemies immediately know where you are. This really forces the player into a more limited style. If I snipe an enemy from really far away and behind cover all of his friends shouldn’t just turn towards me and start shooting. If I melee insta-kill an unaware enemy from cover I should be able to move along and try for his buddy too rather than immediately being discovered.
You could play with visibility via lighting. Certain areas can be dark, or maybe the player can turn off the lights (either by shooting them or by using a non-combat ability like electronics to short a fuse box). This would force the player and the enemies to deal with the darkness somehow. Certain enemies could be effectively blind in the dark or use flashlights (which give away their position), while more advanced or specialized enemies have night-vision capabilities and suffer no difficulty. The player would have more options: helmets could have an attribute that grants/doesn’t grant night vision, or they could have built-in flashlights on either weapons or armor (or they could be add-ons). If you wanted to forgo those options you could fight simply by the muzzle flash of the enemies’ weapons.
Also, the size of the fighting areas should really vary more (in terms of both absolute size and of the “practical” size based on density/frequency of line-of-sight blockers and/or cover). Don’t be afraid to make a claustrophobic level where snipers would have a hard time; similarly you can add more wide-open levels where shotgun/melee characters may be in trouble.
Not every level needs to be of the same -or even similar- difficulty for every group composition. They only need to be reasonably doable. Most of the time the ideal group should be something you can determine logically based on the description of the mission: i.e. you would want to bring a krogan or other heavy-armored close-quarters characters to assault an office building or a lab, but if you need to hold an elevated position you would bring snipers. You can bring a non-optimal group, but you should have a harder time of it. Similarly, enemy composition should also play a large role. Because of their skill sets, certain characters simply destroy geth; but IMO the geth could be a little bit stronger against non-optimal groups to balance this out.
Don’t be afraid to force the player into running and gunning every once and a while. Not every single battle has to be in an area littered with (or even containing any) cover.
With regards to melee combat, I think a lot could be done to expand this into a more useful skill. It should be useful, but only when it makes logical sense for it to. Uncharted 3 does melee really well IMO. Running around a corner right into an enemy should give the enemy a chance to go for a melee attack that you have to counter, rather than you just immediately being able to shoot them.
Movable / destroyable environments really need to play a bigger role. You shouldn’t be able to hide behind a packing crate as three guys shoot rockets at you. By making the battlefield modifiable you allow the battles to be more dynamic. Rather than just move from box to box, cover to cover until you reach the end of the room you can blow open a new route that gives you some advantage. Similarly, you would no longer be able to simply sit behind cover and kill all the enemies in the room if they can blow up all your good cover with a grenade or rocket. This also lets you give combat-specced characters a “non-combat” skill. Maybe a door can’t be hacked, but a sufficiently-heavy weapon can blow it open.
Ammo-powers should be changed to a researched option for weapons, something to be set on weapon-loadout screen before a mission (this forces more consideration to be given before mission: ie if you are expecting geth use anti-synthetic or anti-shield ammo). This would also increase the effect of coming upon an unexpected enemy type (ie finding a biotic on a geth ship).
Armor
The armor system from ME3 is very nearly what I would consider an ideal system. In my opinion, nearly all of the problems are caused by the parts themselves rather than the system. Luckily this can be addressed simply by adding more and varied options for each component slot. But in order for these new components to be anything more than a slightly larger bag of eye-candy for the player, the magnitude of effects offered must be greatly increased.
Instead of values < 10% which often translate into < 1 more survived hit, or killing an enemy in 29 hits rather than 30, single stats like health, shields, and damage need to give 15-20%, enough to actually let the player immediately FEEL the difference the very next time they go into battle. However, if you only have + health, or +shields, or +damage, it begins to feel just like a numbers game; so a wider variety of bonuses are required, preferably additional effects or extra abilities rather than just percentages. For example: you could put night-vision on a helmet, or a chest-piece could have a special usable skill “hardened shield” that grants temporary invulnerability, or gauntlets that give a special longer-reach melee attack, etc. These should also make sense in relation to the piece of armor they are on: sprint-related bonuses should only really be on leg armor, while melee or “strength” related abilities should stay on the arm / shoulder pieces. In my opinion, armor should give you a degree of character customization that is at least as extensive as the skill tree itself.
Also, each piece should be able to be colored individually instead of all armor using the same color scheme. Also, along with more pieces there should be more color/texture patterns available to the user for customization. Some (not necessarily all) should have multiple colors to customize (for example camouflage could have 2 additional user-selectable colors in addition to the base armor color). Any icons or logos also need to be modifiable rather than fixed to the armor.