Knowledge & Context
Tell Vulnix how your application works once, and every future run starts with it.
The free-text context and reference documents on a scope apply to one pentest. Knowledge is the version that sticks: you write something once, and every run it applies to starts already knowing it.
Find it under Knowledge in the sidebar.
What belongs here
Anything true about your application that an agent can't work out by looking at it from the outside, and that will still be true next month. Each entry has a type:
Prop
Type
Accepted Risks is the one people skip and shouldn't. Without it, every run re-discovers and re-reports the same issue you already decided to live with.
Scope
An entry can apply to specific domains and repositories, or to everything.
Leave the scope empty and it applies to every pentest in your workspace — that's the right choice for facts about your organization ("all our APIs require a tenant header"). Pick targets when the fact is about one application and would be wrong or noisy applied to the others.
There is no step for this in the pentest wizard
Knowledge is applied by matching a run's targets, automatically. The wizard shows a line confirming it's in play — you never select entries per run, because that would defeat the point of writing them once.
What Vulnix learns on its own
The Learned tab is the other half: as the agent tests, it takes notes, and the ones describing your application rather than the run in progress are kept when the run finishes. Over several pentests this builds up an understanding of your system that carries into every later run.
These entries can't be edited — the agent wrote them from what it actually observed, and editing one in place would make the page claim something no run concluded. If one is wrong or no longer true, delete it. A wrong entry is worth removing promptly, because it steers every run after it.
Pentests only, for now
PR reviews don't yet contribute to the Learned tab. Only pentests do.
How it reaches the agent
At the start of a run, every entry matching that run's targets is grouped by type and included in the agent's instructions, alongside the per-run context you set in the wizard. Your per-run context comes first — it's the more specific instruction for that particular pentest.
