National Cyber Warfare Foundation (NCWF)login

Pentest-Service-Enumeration for service enumeration cheat sheets and AI endpoint testing


0 user ratings
2026-11-17 03:29:25
milo
Red Team (CNA)
"Pentest-Service-Enumeration

Pentest-Service-Enumeration is a Python reference tool that organizes per-service enumeration commands and adds fingerprinting of AI/LLM endpoints for authorized penetration testers and certification students.








Toolssstonebraker/Pentest-Service-Enumeration — a cheat-sheet lookup for per-service pentest commands plus AI/LLM endpoint fingerprinting and interaction
CategoryEnumeration assistant / AI security testing toolkit
Primary UseLooking up which commands to run against discovered services via pse , and fingerprinting AI endpoints with pse fp during authorized engagements
Safe UseIntended for authorized penetration tests, OSCP/OSWA/OSAI study labs, and internal engagements against systems you own or have written permission to test
Telemetry NoteThe tool performs full-port nmap scans and sends chat/injection requests to target endpoints; its own activity is logged locally under ~/pentest/lab//, and AI endpoints will show its probes in their request logs

ssstonebraker/Pentest-Service-Enumeration, invoked as pse, addresses a problem every operator knows intimately: the enumeration phase produces a growing list of services, and for each one there is a fog of commands, tools, and flags half-remembered from previous engagements. The author, a working penetration tester who built the tool while studying for the OSCP and OSWA certifications, originally framed it as a personal command ledger and later expanded it into a full enumeration assistant. With 119 stars, a Python codebase requiring 3.10+, and support for Linux, macOS, and WSL, it sits squarely in the practical, operator-built corner of the tooling ecosystem rather than the polished-commercial one.


The core lookup interface is elegantly minimal. Typing pse renders the stored cheat sheet for that service, while pse 445 or pse 53/udp resolves a port and protocol back to the services that typically live there. pse -l lists everything available, pse -r "nmap" searches across all entries, and -e, -c, and -d provide edit, create, and delete operations on the service files themselves. The design bet here is that your enumeration notes should be versioned, searchable, and editable in-place rather than scattered across markdown files, OneNote pages, and shell history — and the port-based indirection is a genuinely useful touch when you are staring at an nmap output and thinking in ports, not service names.


The shipped service catalog is visible in the README's usage output and covers the standard internal-network surface: smb, kerberos, ldap, adcs, dns, mssql, nfs, rpc, snmp, smtp, ssh, ftp, and http/web, plus utility entries for nmap workflows, mimikatz module references, Linux privilege escalation checks (linpriv), and web fuzzing with wfuzz. Each entry carries its associated ports — adcs maps to 389/tcp, 636/tcp, and 88/tcp, for example — and a one-line description of what the stored commands do, such as dumping certificate authorities and templates or enumerating domain information with ldapdomaindump. The files live under /root/.pse, which means the whole knowledge base is plain, user-editable data rather than a compiled binary blob.


What elevates PSE beyond a glorified cheat clone is the AI reconnaissance layer the author bolted on after AI-powered services started appearing on client engagements. pse fp performs a fingerprinting pass that the README describes as covering all 65535 TCP ports via nmap plus AI-specific probes, with a --fresh flag to bypass caching, --ports to probe a shortlist of specific ports in roughly five seconds, and --targets for batch fingerprinting of CIDR ranges with concurrency capped at four. When it detects an Ollama or Open WebUI endpoint, it also enumerates available models — visible in per-port output without any credentials — which is exactly the kind of surface mapping an assessor needs before deciding how an exposed LLM deployment fits into scope.


The pse chat command provides terminal interaction with discovered AI endpoints, including an interactive REPL mode and credential-based access to Open WebUI instances via --creds. On authentication it auto-discovers the available models and configured tools, selecting the first model unless overridden with --model. The README documents two tool-execution modes that matter for understanding the architecture: the native mode, where tool specifications are passed to the LLM and tool calls come back to the client without server-side execution, and the --legacy-fc mode, where the server makes a second LLM call to resolve parameters and executes the tool inline before responding. That distinction — observe versus execute — is a meaningful one for anyone assessing MCP tool integrations, and the README is explicit that legacy function calling is the path of interest when evaluating whether tool servers can be abused.


This is unambiguously offensive-capability territory, and the framing matters: the playbook system, the pse inject web-content injection feature, the pse review code-review-agent module, and the pse rag retrieval-hijacking mode described in the README's table of contents are all designed for use inside authorized engagements, certification labs, and controlled test environments. Everything a tester does through the chat interface is logged to ~/pentest/lab//chat/ with full curl logging for report reproduction, which is a professional touch — the tool is built to support deliverable writing, not just hitting targets. Credentials can be supplied through the PSE_CREDS environment variable specifically to keep them out of ps output, another sign of an author who thinks about operational hygiene on real assessments.


The broader feature surface sketched in the README reads like a catalog of emerging LLM attack classes. Beyond the MCP tool work there is a multi-technique extraction engine with adaptive concurrency governed by an AIMD scheme, an indirect injection workflow under pse inject, a code-review exploitation mode under pse review, RAG retrieval hijacking under pse rag with auto-escalation and output-encoding options, a plugin architecture for custom decoder techniques, an evidence mode, an embedded MCP server, and a bundled Grit pre-checkout security scanner. The README also mentions pse fp --creds for authenticated Open WebUI tool discovery and --token for GitLab-authenticated scanning, indicating the fingerprinting layer respects authenticated contexts where relevant. Notably, the author flags the whole AI module as useful preparation for the OSAI certification, situating it in the training ecosystem as much as the field toolkit one.


Architecturally, the split between a static, file-backed cheat-sheet engine and a dynamic network client is worth appreciating. The service catalog is declarative data under ~/.pse, creatable through pse -c templates and editable with pse -e, which invites teams to maintain a shared, customized knowledge base in a repository. The AI subsystem, by contrast, is stateful: sessions persist (pse --reset starts a new chat context, pse --status reports the current one), and every interaction lands in timestamped log files. That separation keeps the low-risk lookup functionality fast and offline while isolating the network-touching features behind explicit subcommands.


Installation follows the conventional pattern: git clone https://github.com/ssstonebraker/Pentest-Service-Enumeration, then ./install.sh from the repository root, after which the pse entry point is available. The license is marked NOASSERTION in the repository metadata, though the README displays a "Sustainable Use" badge, so teams intending to bundle or redistribute it should read the LICENSE file carefully before relying on it in commercial tooling. The requirement of Python 3.10+ and the Linux/macOS/WSL platform support keep it aligned with typical pentest workstation setups.


For defenders, the telemetry angle is straightforward but worth stating: pse fp in full mode is a complete TCP port scan and will look like one on the wire, while the AI probes, chat sessions, and injection tests generate distinctive HTTP request patterns against LLM endpoint paths. If you run exposed Ollama or Open WebUI deployments in your environment, this tool is a useful reminder that model enumeration, tool discovery, and authenticated chat interaction against those services require no browser and leave clean client-side logs on the tester's side — meaning your server-side request logs are the primary detection surface. Reading this repository is a compact education in how AI infrastructure is being assessed in 2026, and for authorized testers prepping for OSCP or OSAI, the service catalog alone justifies a bookmark.



Official project repository for ssstonebraker/Pentest-Service-Enumeration.

Download Tool

Educational analysis for authorized security professionals. Use only in controlled, authorized environments.






Source: OffensiveSec
Source Link: https://www.offsecblog.com/2026/11/pentest-service-enumeration-for-service.html


Comments
new comment
Nobody has commented yet. Will you be the first?
 
Forum
Red Team (CNA)



Copyright 2012 through 2026 - National Cyber Warfare Foundation - All rights reserved worldwide.