confirmesc

A Linux privilege-escalation scanner that confirms what is actually exploitable
and hands you the exact command to become root.

powered by tmrswrr

Why  •  Demo  •  Quick start  •  AI / MCP  •  What it checks  •  Usage

---

Most enumeration tools dump hundreds of "maybe interesting" lines and leave the hard part to you: figuring out which ones are real, then searching the web for a working exploit.

confirmesc does the opposite. It verifies each finding directly against the live system, labels it with an honest confidence level, and for every confirmed vector it prints a ready-to-run command that actually gets you a root shell. No guessing which of 200 lines matter, and no opening GTFOBins in another tab to look up the payload.

Built for pentesters, CTF / Hack The Box players, and system administrators who want a clear answer to one question: can this box be escalated right now, and how?

Why confirmesc is different

	Classic enumerators (LinEnum, linPEAS)	GTFOBins (reference site)	confirmesc
Lists suspicious items	Hundreds of lines	-	Only what matters
Confirms it is exploitable on this box	No	No	Yes, verified live
Gives the exact exploit command	No	Look it up by hand	Yes, ready to paste
Handles the dash/bash privilege drop	No	Often fails	Yes
Proves root for real (euid == 0)	No	No	Yes, with --poc
Honest confidence levels	No	-	CONFIRMED / LIKELY / INFO
AI agent integration (MCP + skills)	No	No	Yes

Five things set it apart:

• It confirms, it does not guess. A finding is marked CONFIRMED only when the exploitable condition itself was directly observed on the running system: a real SUID bit on a known binary, a real dangerous capability, an authorized sudo -l rule, a genuinely writable root-owned file. Not "this version is probably vulnerable."

• It gives you the payload, so you never search for it. Every confirmed SUID / sudo / capability vector ships with a copy-paste command that spawns an interactive root shell. No opening GTFOBins, matching the technique, and adapting a one-liner by hand - confirmesc already did that.

• Its commands work on modern boxes. On current Debian-based systems, /bin/sh is dash, and both dash and bash silently drop root the moment they notice euid != ruid. Many copied GTFOBins one-liners fail because of this. confirmesc generates commands that account for it (explicit setuid(0), bash -p, or a direct exec()), so the shell you get is really root.

• It can prove root, live. With --poc it runs the real escalation payload and checks that the resulting process truly has euid 0, turning "exploitable in theory" into "we got root just now."

• It is AI-agent-ready. confirmesc ships an MCP server and a set of skills, so an AI agent can run the scan as a tool and follow a built-in methodology to act on the confirmed findings - AI-assisted privilege escalation, end to end. See AI agent integration.

Demo

An unprivileged user runs confirmesc, gets three CONFIRMED vectors each with a ready-to-run root shell, pastes one, and lands a root shell - recorded live in the testlab/ Docker sandbox.

Quick start

git clone https://github.com/capture0x/confirmesc
cd confirmesc
python3 -m pip install -e .

confirmesc            # passive scan of the current box, readable text report

Read the top of the report. The ACTIONABLE FINDINGS section lists every CONFIRMED vector, each with a highlighted root shell: line you can copy and run. That is the whole workflow: scan, read the confirmed findings, run the command.

The process exit code is 1 when any CONFIRMED finding exists and 0 otherwise, which makes it easy to use in CI or CTF automation.

AI agent integration (MCP + skills)

confirmesc is built for AI-assisted privilege escalation. It ships two optional pieces that let an AI agent drive it over the Model Context Protocol: the scan becomes a callable tool, and a set of skills gives the agent a built-in methodology to act on the confirmed findings. The agent runs the scan, reads the confirmed vectors, and already has the exact root command for each one - no manual triage, no searching for payloads.

MCP server

confirmesc-mcp exposes scanning over the Model Context Protocol. An agent lists the tools, calls scan, and gets structured findings back, each CONFIRMED one already carrying its ready-to-run exploit command:

Tools exposed:

• scan - run a passive/active scan and return structured findings (with the exploit_command for each confirmed vector)
• report - run a scan and return a rendered report as text, json, or html
• list_privesc_checks - list the available checks by category

Live exploitation (--poc) is deliberately not exposed: confirming conditions is safe to automate, but running escalation payloads stays an operator decision.

pip install -e ".[mcp]"
confirmesc-mcp            # serve over stdio

Point any MCP client at it with a standard stdio server config:

{
  "mcpServers": {
    "confirmesc": { "command": "confirmesc-mcp" }
  }
}

Skills

The skills/ directory holds playbooks an AI agent loads to work through each vector class. There is one skill per confirmesc finding type, plus an orchestrator: