Tools & runtime extensions
Learn about the tools available during public beta, including scoped Browser/HTTP and optional offline Docker. Shell, WSL and MCP execution are not yet available.
Tool categories
| Category | Examples | Execution boundary |
|---|---|---|
| Built-in desktop tools | HTTP requests, Browser Agent, code grep; native shell currently blocked in packaged beta. | Desktop runtime with beta restrictions. |
| Kali knowledge | tools.search, tools.help. | Read installed command names/help from the desktop/WSL environment. |
| Custom executable tools | Saved command templates with named placeholders. | Rendered into a normal shell.command path. |
| MCP | mcp.listServers, mcp.listTools, mcp.call. | Connected stdio child process in Electron main. |
| Cloud execution | Disabled during the public beta; this is separate from Cloud AI chat. | Unavailable until release approval. |
| Specialist delegation | recon, web, auth, injection, verification. | Isolated model analysis; no direct execution bypass. |
Kali tools and WSL availability
PENTER does not inject thousands of tool manuals into every prompt. Agents can first search the installed command catalog, then read exact local help/man output for the command they actually intend to use.
tools.search({ query: "ldap" })
tools.help({ command: "ldapsearch" })Create custom executable tools
Open the Extensions/Plugins view and add a custom tool with a name, description, command template and optional tags. Placeholders use braces and become named arguments that the agent supplies through customTool.run.
nuclei -u {url} -tags {tags}customTool.run
toolId: my-nuclei
arguments: { "url": "https://authorized.example", "tags": "exposure,misconfig" }- Missing placeholders cause the custom tool call to fail instead of silently rendering an incomplete command.
- Newline and NUL characters are stripped from placeholder values before substitution.
- Disabled custom tools disappear from the enabled tool list exposed to agents.
- The final rendered command goes through the normal shell-command execution/approval path.
Background-capable tool work
PENTER exposes a background-tool wrapper for selected non-shell actions. The current supported set includes HTTP request, Browser Agent action, code grep and web search aliases. This lets the conversation continue while a supported non-shell task is running.
Why shell scanners are different
Extensions do not bypass approval
Custom executable tools eventually become a shell command. MCP tool calls have their own explicit approval binding. Starting an MCP server is also approval-gated. This keeps the extension layer from becoming a side door around PENTER's normal operator-control model.
- Review command/tool name and arguments before granting approval.
- Prefer one narrow extension over a general-purpose wrapper when possible.
- Keep extension credentials in the extension/MCP environment rather than embedding them in a command template.
- Disable or remove extensions that are not needed for the active engagement.
Ready to get started?
Follow the setup guide, install the current approved Desktop release, and connect your account.