The Basic Set cautions against permitting Ninja Talent and Weapon Talent, which some gamers read as a ban on Talents benefiting combat skills.
There’s no such prohibition, though!
Ninja Talent is singled out not because it might help combat skills, but because ninja training is extremely broad; outside of a silly or cinematic campaign, it’s implausible that anyone would be gifted at everything involved (but see Talents as Training, pp. 25-26, and note that there’s a Ninja Talent of sorts on p. 7).
Weapon Talent is likewise flagged because it covers too much – in the typical action-adventure campaign, it would usurp DX.

　A balanced combat Talent meets three criteria:

　1. It encompasses a small number of combat skills – probably fewer than five, discounting individual specialties.
For example, Hot Pilot (p. 11) and Seafarer (p. 14) have one combat skill apiece; Strangler (p. 15) affects two; Pickaxe Penchant (p. 14), three; and Halfling Marksmanship (p. 11) and Mr. Smash (p. 13), four each.

　2. The affected combat skills are closely related in the motions or thinking involved, making the Talent a credible aptitude. Mr. Smash benefits two-handed swings, for instance, while Halfling Marksmanship helps one predict the arc of light, low-speed projectiles. Combat skills that default to one another are often related.

　3. It includes noncombat skills that fit the concept; e.g., Mr. Smash aids Forced Entry because it’s about smashing things. “Diluting” combat skills with noncombat ones is an entirely reasonable way to increase a Talent’s cost enough that players select it to aid characterization, not to save points.

　Broad combat Talents are questionable and open to abuse.
In particular, skills related only because they involve violence and are often learned together – like those in a fighting style from GURPS Martial Arts or the template for a warrior archetype – form a poor basis for a Talent, as they’re linked only by the whims of a teacher or a job.
For outright dissimilar skills like Gunner, Knife, and Polearm, Talent definitely isn’t believable.
Such things occasionally work as Job Training (pp. 25-26), or as “racial” Talents (pp. 4-5) programmed into golems or robots, but not as aptitudes.

　The GM should fight the urge to forbid combat Talents that do meet the listed criteria, though.
Even in an actionadventure story, combat isn’t everything – and no player bent on “gaming the system” will buy Talent for 5-15 points/level if he can raise his PC’s favorite violent skill for 4 points/level!