NullSquare

Watch Agent

Continuous CVE exposure: match each new advisory against the software you actually run, try to reproduce it, and say so either way.

A new CVE is published. The only question that matters is whether the affected version runs somewhere you own. The Watch Agent answers that question against your asset inventory, and then tries to reproduce the issue before it calls it an exposure.

It reports the all-clear as well. An agent that speaks only when it has bad news gives you no way to tell quiet from broken.

What the agent does

The Watch Agent reads new advisories continuously and compares each one against the software the Pentest Agent has already observed on your assets. It never matches on your industry or your vendor list. It matches on the component and the version it read from your surface.

A match is a reason to look, not a result. The agent then attempts the documented vector against the affected asset, inside the same rules of engagement the scope sets for every other assessment.

  • It watches the assets already in your inventory. Nothing new to configure.
  • It matches on observed component and version, never on sector or guesswork.
  • It stays inside the scope. The rules of engagement apply here as everywhere.

The four states

Every advisory the agent raises carries one state, and the state says exactly how much is known. The distinction between "we found a version match" and "we made it happen" is the whole point of the agent, so the two never share a word.

A confirmed exposure. The reasoning panel shows what the agent read and why it matched; the outcome line records what it managed to reproduce.
  • Potential — the version matches an advisory. Nobody has tried it yet.
  • Verifying — the agent is attempting the documented vector right now.
  • Confirmed exploitable — the agent reproduced it against your asset.
  • Not reproduced — the agent tried and failed. The version matched; the exposure did not.

Why the agent thinks it applies

Every advisory shows its own reasoning before it shows its verdict. You can see the component the agent observed, the identifier it resolved it to, the basis of the match, and when it last saw the component on that asset.

This is what lets you disagree with the agent. A backport risk or a version conflict is flagged as such, because a distribution can patch a component without changing the version string the agent can read.

The reasoning is shown before the verdict, so a match you disagree with is one you can argue with.

How exposures are ordered

Severity alone does not tell you what to do first. Each exposure carries the CVSS score from the advisory, the EPSS percentile for how likely exploitation is in the wild, and whether CISA lists it as known-exploited. An exposure the agent reproduced on an asset you own outranks a higher CVSS the agent could not reproduce.

  • CVSS — how bad the issue is, according to the advisory.
  • EPSS — how likely it is to be exploited, as a percentile.
  • CISA KEV — whether it is known to be exploited in the wild.
  • Reproduced — whether it works here, which outranks all three.

Where exposures appear

Exposures are risks, so they arrive in the same inbox as everything else. Risks holds what the Pentest Agent proved, what the Code Review Agent found in a pull request, and what the Watch Agent reproduced from an advisory. One list, one triage rhythm, one place to assign an owner.

One inbox. CVE exposures and pentest findings are both risks, and both are triaged the same way.

Related articles