Using a specialised tool is probably the way to do this reliably.
However, one really janky way of doing this would be to grab a list of all the .class files in your JAR, put the JAR on the classpath and use javap to get references to other classes:
#!/usr/bin/env bash
javap -cp $1 -v \
`zipinfo -1 $1 '*.class' | sed 's:/:.:g' | sed 's:\.class$::'` | \
grep ' = Class' | sed 's:.*// ::' | sort | uniq
Running this on guava-19.0.jar gives this:
"[[B"
"[B"
"[[C"
"[C"
com/google/common/annotations/Beta
com/google/common/annotations/GwtCompatible
com/google/common/annotations/GwtIncompatible
com/google/common/annotations/VisibleForTesting
com/google/common/base/Absent
com/google/common/base/AbstractIterator
...............................................................
"[Lcom/google/common/util/concurrent/MoreExecutors$DirectExecutor;"
"[Lcom/google/common/util/concurrent/Service$State;"
"[Lcom/google/thirdparty/publicsuffix/PublicSuffixType;"
"[Ljava/io/File;"
"[[Ljava/lang/annotation/Annotation;"
"[Ljava/lang/annotation/Annotation;"
"[Ljava/lang/Class;"
"[Ljava/lang/Comparable;"
"[Ljava/lang/Enum;"
"[[Ljava/lang/Object;"
"[Ljava/lang/Object;"
"[Ljava/lang/reflect/Field;"
"[Ljava/lang/reflect/Method;"
"[Ljava/lang/reflect/Type;"
"[Ljava/lang/reflect/TypeVariable;"
"[Ljava/lang/StackTraceElement;"
"[Ljava/lang/String;"
"[Ljava/net/URL;"
"[Ljava/util/Iterator;"
"[Ljava/util/Map$Entry;"
"[[S"
"[S"
sun/misc/Unsafe
"[[Z"
"[Z"
You'll need more output formatting, and, as others have pointed out, it won't pick up any use of reflection.
How this works:
zipinfo -1 $1 '*.class' will print out the names of all .class files in $1, which is the argument to the script shown. The seds change /s to .s and remove the .class extension, so that you end up with a list of Java-style class names. You could do this more elegantly, but it should work.
The javap invocation puts the jar on the classpath with -cp, and passes all the classes. -v makes it output a lot of information, including some entries which represent references to names of classes. The grep ensures we're only looking at those, the sed removes some extra information we're not interested in. sort | uniq ensures we're not printing the name of any class more than once. It does need a bit more sedding to standardize an output format.