How Signal works 1.0

How Signal works

آخر تحديث 10 نوفمبر 2026
في هذه الصفحة

From the records Windows keeps to a short list of signals: the five steps of every run, and the ideas behind them.

The five steps

  1. 1. Read what Windows recorded

    Signal reads the host's own records: event logs, the registry, the service and process tables, scheduled tasks, Prefetch, browser databases, antivirus status and more. Each source is read once per run, and shared by every module that needs it.

  2. 2. Turn records into evidence

    Each record becomes a piece of evidence of a known type - a process_creation, a logon, an autostart entry - with named fields such as process.name or account.sid. The same thing seen in two places (a service in the System log and in the service table) is merged into one record that names both sources.

  3. 3. Run the rules

    Each module runs its rules over the evidence. A rule fires when its condition is true for a record (a shell started by Word) or for the run as a whole (no audit policy enabled at all).

  4. 4. Group matches into signals and entities

    Every match of a rule about the same thing becomes one signal, with a count and its first and last time. Signals about the same program, account or service are grouped into one entity, across modules.

  5. 5. Say what could not be seen

    Signal records the health of every source: read, partly read, not read and why. A rule whose data was missing is not run - never "clear". This is the "what you can't see" half of every run.

Modules: one question each

Signal's work is divided into modules. Each module answers one question an analyst asks about a host:

Module

The question

Executions

What ran on the host, and how was it abused?

Persistence

What will run again without anyone starting it?

Logons

How was the host used, and from where?

Users

Who used the host?

LogIntegrity

Can the host's logs be trusted?

Protection

What protects the host from malware, and does it work?

Network

What does the host listen on and talk to?

RemoteAccess

Which remote access tools could reach and control the host?

Browsers

What came in through a browser?

Webshells

Which web sites does the host serve, and do they hold a webshell?

Applications

What software is installed?

Backup

Can the host be recovered, and did someone take that away?

Directory

What did the domain push to this host through Group Policy?

Scope

Is this host in scope of the incident?

The ideas behind it

Signals, not inventories

Signal does not dump every process, every logon or every command line for you to read. It outputs what the rules matched, plus small capped tables. You get a list to act on, not a new pile of data to analyse.

One thing, one record

A user, a process, a service or a task is read from every source that records it and merged into one record, with each part attributed to its source. An account is matched by its SID, never by its name alone.

Unknown is never "no"

When Signal cannot read something, it says so. A log that does not exist is absent; a rule that could not be evaluated is not run. Silence in Signal's output never means "nothing happened" unless the source was actually read.

Never disrupt the host

Signal only reads. Locked files are read without taking locks; no process is stopped; no shadow copy is ever created. It runs at below-normal CPU priority, so the host's own work comes first.

Next

See where Signal fits beside the other tools of incident response.