Currently the XSSAuditor has very few false positives, but it does have a number of false negatives. False negatives a problematic because they let attackers bypass the filter, negating some of its security benefits. False positives fall into two categories:
1) Tricky injections into in-scope injection contexts.
2) Injections into out-of-scope injection contexts.
Whether an injection context is in-scope is somewhat of an arbitrary distinction, but we should aim to tackle all the tricky injection vectors for the in-scope injection contexts before we expand to consider more contexts in-scope. This approach puts us on a path to eliminating all the attack vectors for a given set of vulnerabilities rather than catching some, but not all, the attack vectors for a large number of vulnerabilities.
Let's use this bug as a meta bug for Type 1 vulnerabilities (which will appear as blocking this bug).