PenterDocumentationDesktop workspace
Documentation

Desktop workspace

PENTER Desktop is a security workbench rather than a chat window. The agent, target state, browser, execution queue, files and findings are meant to remain connected to the same investigation.

01

Treat the workspace as the engagement boundary

The workbench is designed around persistent context. Instead of copying data between a terminal, proxy, browser, notes app and AI chat, PENTER keeps the selected target and execution history close to the current agent thread.

Do not reuse one workspace as a global scratchpad
Mixing unrelated customers or targets makes approval decisions, saved browser state, extension configuration and evidence provenance harder to reason about. Start a separate workspace when the authorization boundary changes.
02

Workbench surfaces

Agent

Plan, investigate, request tools, inspect results and continue the technical reasoning.

Target explorer

Keep the active target, browser context and pinned information visible to the workflow.

Findings

Review candidate issues, verification state, evidence and reporting readiness.

Tool catalog

Browse capability-oriented security tooling without forcing every tool manual into the model context.

Terminal / Queue

See proposed/running command work, inspect output and stop work when the runtime supports cancellation.

Files

Keep notes, PoCs and engagement artifacts beside the investigation.

Inspector

Show contextual details for the selected target, tool, finding or workbench item.

Extensions

Manage custom command tools and MCP server integrations from the Plugins/Extensions view.
03

What belongs in engagement state

StateWhy it mattersRecommendation
TargetGives browser, agent and Cloud execution a common scope anchor.Set it before active testing.
Browser stateAuthentication/navigation state changes what the application exposes.Keep state tied to the same engagement.
FindingsCandidates need provenance and verification context.Do not promote scanner noise automatically.
Tool resultsResults feed subsequent reasoning.Keep raw output available while summarizing only what is needed.
Files / PoCsArtifacts often outlive a single agent turn.Store them inside the engagement workspace.
ExtensionsMCP/custom tools can grant powerful capabilities.Enable only what the engagement actually needs.
04

Execution should stay inspectable

PENTER separates model reasoning from execution. A tool request should produce a visible action or result row, and approval-sensitive operations are checked by the desktop execution layer rather than being silently treated as already complete.

  • Review the actual command or MCP tool before granting approval.
  • Use the queue/terminal output as evidence of what executed rather than relying on conversational claims.
  • Keep active target context visible when reviewing commands that can affect remote systems.
  • Use Cloud execution only when its stricter target and command policy matches the intended action.
05

Files, findings and evidence

The workspace should make it possible to move from observation to verification without losing the path that produced the issue. Notes, screenshots, request/response evidence and PoCs are more useful when they remain tied to the finding they support.

A good finding is reproducible
Prefer a concise reproduction path, technical evidence and impact explanation over a raw scanner label. Keep candidate observations separate until the evidence supports the claim.
Next step

Ready to get started?

Follow the setup guide, install the current approved Desktop release, and connect your account.