From a sentence to a signed pentest.In eight autonomous steps.

1. // 01

   Describe the target in plain English

   Tell DXSense what you want tested. Scope. Out-of-scope. Depth. Constraints. The platform parses the goal. If anything is ambiguous it asks a clarifying question before a single packet leaves the sandbox.

   WhyNatural-language intake removes the config-file ceremony that slows down traditional engagements. The clarification loop prevents over-reaching your authorisation.

2. // 02

   A plan is synthesised

   Your goal is decomposed into a phased engagement: reconnaissance. Vulnerability validation. Exploitation. Privilege escalation. Post-exploitation. Reporting. You see the full plan before anything runs.

   WhyPlanning as data beats a black box. You can audit. Edit. Approve the engagement plan before execution. The difference between an assistant you trust and a rogue tool you do not.

3. // 03

   You approve the plan

   Nothing executes until an operator explicitly approves. You can prune steps. Edit scope. Request an alternative path. Human-in-the-loop gates are wired at every destructive or privilege-changing step. Not just at the start.

   WhyAutonomy without authority is reckless. Approval gates are how we keep the system insurable. Auditable. Inside the rules of engagement.

4. // 04

   Autonomous execution

   Specialised capabilities execute the engagement in parallel where safe. Sequentially where dependent. New specialised workers are synthesised on demand when a sub-goal needs a skillset the default roster does not cover. Every worker has a narrow testable responsibility.

   WhyNarrow focused workers beat a monolithic super-tool. Reasoning stays focused. Failure modes stay isolated. The whole system stays legible to a reviewer.

5. // 05

   Tools run in an isolated sandbox

   Industry-standard offensive tooling runs inside isolated sandboxes. Every command. Every argument. Every exit code. Every artifact. Captured. No exploit is hallucinated. Every one is actually detonated then verified.

   WhySandbox-verified execution is the line between a security report you can act on and a language model's confident guess. If it is in the report it ran.

6. // 06

   Findings and sessions are captured live

   Shells. Credentials. Privileged sessions. Recorded with timestamps. Provenance. Replayable PoCs. The plan adapts on every finding. If SQL injection lands exploitation is queued. If a shell drops privilege-escalation is re-weighted.

   WhyAdaptation is what separates agentic AI from a scripted runner. The plan is a living graph. Updated on evidence. Never on assumptions.

7. // 07

   Evidence is sealed and signed

   Every event. Every artifact. Every session log. Cryptographically sealed. Signed with your operator key. The report ships with the signature. The verification root. A verification script auditors can run independently.

   WhyTamper-proof evidence turns a pentest report from a PDF of claims into a cryptographic artifact regulators. Auditors. Courts can verify for themselves.

8. // 08

   You review. Export. Hand off.

   Review the engagement in the dashboard. Export a signed PDF. Hand the evidence chain to SOC or GRC or legal. Every action is replayable. Every action is attributable.

   WhyPentesting value is not the exploit. It is the remediation workflow. Signed structured output plugs into the stack you already use.

// OPERATING PRINCIPLES

The rules the platform lives by.