You could use a paste approach with the help of gsub:

A <- c("ok","WA","WA","ok","WA")
B <- c("WB","ok","ok","ok","WB")
C <- c("WC","ok","WC","ok","WC")
df <- data.frame(A=A, B=B, C=C, stringsAsFactors=FALSE)

df$D <- paste(df$A, df$B, df$C, sep=",")
df$D <- gsub("^,|,$", "", gsub(",{2,}", ",", gsub("\bok\b", ",", df$D)))
df

   A  B  C        D
1 ok WB WC    WB,WC
2 WA ok ok       WA
3 WA ok WC    WA,WC
4 ok ok ok         
5 WA WB WC WA,WB,WC

The basic strategy here is to strip out ok entries, then cleanup the possible dangling commas left from the paste call.

Using apply and paste

df['D'] <- apply(df, 1, (x) paste(x[which(x!="ok")], collapse=","))
df
# A tibble: 5 × 4
  A     B     C     D         
  <chr> <chr> <chr> <chr>     
1 ok    WB    WC    "WB,WC"   
2 WA    ok    ok    "WA"      
3 WA    ok    WC    "WA,WC"   
4 ok    ok    ok    ""        
5 WA    WB    WC    "WA,WB,WC"

A non-vectorised alternative using applyrow-wisely and toString with nchar. This might be fast enough on moderate sized data.

df0 = data.frame(A = c("ok","WA","WA","ok","WA"), 
                 B = c("WB","ok","ok","ok","WB"), 
                 C = c("WC","ok","WC","ok","WC"))

df0$D = apply(df0, 1, (i) {x = sub("ok", "", i); toString(x[nchar(x)>1L]) })
df0$D[df0$D==""] = "NO W"
df0

# A  B  C          D
# 1 ok WB WC     WB, WC
# 2 WA ok ok         WA
# 3 WA ok WC     WA, WC
# 4 ok ok ok       NO W
# 5 WA WB WC WA, WB, WC

Using dplyr/tidyr, replace “ok” with NA, then unite excluding NA, then column bind to original data.frame:

library(dplyr)
library(tidyr)

df %>% 
  mutate(across(A:C, ~if_else(.x == "ok", NA, .x))) %>% 
  unite(D, A:C, sep = ",", na.rm = TRUE) %>% 
  bind_cols(df, .)

#    A  B  C        D
# 1 ok WB WC    WB,WC
# 2 WA ok ok       WA
# 3 WA ok WC    WA,WC
# 4 ok ok ok         
# 5 WA WB WC WA,WB,WC