I like this as a starting point, but I feel as if all the DCs except the first one ought to be 5 higher. Lucid dreaming is a very rare thing; I've done it precisely thrice, and I was never able to make any changes to the dreamscape during those episodes, so I suspect it's more difficult - knowing you're dreaming just involves becoming aware, while changing anything requires enforcing your will over your perceptions and thus is akin to disbelieving a high-level illusion (with your subconscious as the caster, so it'd be vaguely as if you were possessed by a demon that was casting illusions on your own eyes which it disbelieves while looking out through them).

I also think there's a problem with the fact that the nonlethal damage on awakening is incapable of killing someone, even if they have a heart condition or the like. Use of lucid dreaming as an assassination effort should be almost impossible, but not inherently defined by the system as completely so. I'd go for something like "roll the target's Hit Dice as nonlethal damage; if reduced to 0, the target fails to awaken. Make a Will save 5 minutes later to wake up; if you fail, you slip into a coma. While comatose, make a Fortitude save; on a failure you die." Something along those lines. The probability of death is extremely low, but if you spammed this effect nightly you'd have a chance of it eventually working, especially if you targeted people who appeared frail or weak-willed.

Quote Originally Posted by tuggyne View Post
Secondly, the synergy with Forgery is odd; I would expect Bluff to work better, since you're not writing anything down.
The problem is just that Bluff already has three full synergies and a conditional one attached to it; I'd be happier not having it become more of a god-skill. Perhaps Knowledge: Psionics or Concentration instead?

Quote Originally Posted by nonsi View Post
Many times In literature/movies, getting killed in a manipulated dream is potentially lethal, or at least something that can scar you for life.
To the extent where this is far too much of a cliche, IMO. Reality suggests that there is little evidence for this, and so I disfavor the idea of making it too reliable, as it breaks suspension of disbelief.