|
|
|
@ -81,16 +81,14 @@ def output_to_markdown(detector_classes, filter_wiki): |
|
|
|
|
idx = 1 |
|
|
|
|
for (argument, help_info, impact, proxy, v2) in detectors_list: |
|
|
|
|
print('{} | `{}` | {} | {} | {} | {}'.format(idx, |
|
|
|
|
argument, |
|
|
|
|
help_info, |
|
|
|
|
classification_txt[impact], |
|
|
|
|
'X' if proxy else '', |
|
|
|
|
'X' if v2 else '')) |
|
|
|
|
argument, |
|
|
|
|
help_info, |
|
|
|
|
classification_txt[impact], |
|
|
|
|
'X' if proxy else '', |
|
|
|
|
'X' if v2 else '')) |
|
|
|
|
idx = idx + 1 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def output_detectors_json(detector_classes): |
|
|
|
|
detectors_list = [] |
|
|
|
|
for detector in detector_classes: |
|
|
|
@ -120,7 +118,7 @@ def output_detectors_json(detector_classes): |
|
|
|
|
'impact': classification_txt[impact], |
|
|
|
|
'wiki_url': wiki_url, |
|
|
|
|
'description': description, |
|
|
|
|
'exploit_scenario':exploit, |
|
|
|
|
'recommendation':recommendation}) |
|
|
|
|
'exploit_scenario': exploit, |
|
|
|
|
'recommendation': recommendation}) |
|
|
|
|
idx = idx + 1 |
|
|
|
|
return table |
|
|
|
|