Docs / Run scans
Authorised scope
Each workspace keeps a list of what you may test. Every launch is checked against it before anything reaches a node.
Each workspace keeps a list of the destinations you are stating you may test, under Arsenal, then Authorised scope. Every launch is checked against it before anything is sent to a node: a scan started from the dashboard, each routine stage and mission step, a Bench request, a scan Valkyrie proposes, and a scan an agent asks for through crossfyre mcp. A launch whose target the list does not admit is refused, and nothing is sent. A new workspace starts with the list empty, and while it is empty every launch is refused with a message telling you to add the destinations you are allowed to test.
Only a person adds to the list, and each entry shows who added it and when. Nothing a scan finds is added. An entry is a destination, not a URL, written in one of these forms:
api.example.comThat host, on any port. An address is written the same way.api.example.com:8443That host, only on that port.*.example.comSubdomains of it, not example.com itself.10.0.0.0/8Every address in that range.Paths, schemes and a bare * are refused rather than guessed at.
What a launch needs
- Port scans and crawls can reach any port on their target's host, so they need the host written without a port. A port scan of a range needs one entry that contains all of it.
- Subdomain enumeration needs
*.example.comwhen its active pass is on, since that pass contacts names under the domain. A passive-only run is also admitted byexample.com. - Origin discovery needs both
*.example.comandexample.com. - Every other scan, and a Bench request, goes to one host and port. An entry for the host admits it, and so does one for the host on that port.
Hosts named in a scan's settings are checked too. A saved flow with a step on a host the list does not admit, or a GraphQL endpoint given as a full URL on such a host, refuses the launch. Extra endpoints and takeover hosts the list does not admit are left out, and if none is left the launch is refused. A crawl's extra hosts are left out the same way, its subdomains switch is turned off unless the list covers the names under its seed, and its external links switch is always turned off, since that could reach any host.
Routines and missions pass the hosts one stage finds to the next, and only the ones the list admits go on. A routine stage with none left is skipped, and a mission step with none left fails, so a fail branch can take over. A mission step with targets typed into it is checked like a launch from the dashboard.
When the dashboard refuses a launch, it lists what was not admitted and the entry that would admit each one. If your role lets you change the list, you can confirm that you are authorised to test them, add them from there, and launch again.
The list governs what a launch is aimed at. Once a scan runs, requests an engine makes on its own are fenced only where that engine fences them: the vulnerability engine checks each endpoint's host against the target and the list before it injects, but redirects an engine follows are not checked yet. Removing an entry does not stop a scan that is already running.