Hunt Ics by Encod3d-Sec — AI Skill for Any LLM

Overview

Publisher	Encod3d-Sec
Repository	`TORCH`
Skill name	`hunt-ics`
Stars	334
Forks	48
Bundled files	Instructions only
License	MIT
Links	Skill source Repository

Markdown instructions

A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

Works with any LLM

AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

Self-contained

Everything the model needs lives in the instructions — no extra files to sync.

Open source

Published by Encod3d-Sec on GitHub. Read the source before you install it.

Installation

Install the Hunt Ics AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

Open the app and go to Plugins → Skills.
Choose "Install from GitHub".
Paste the skill folder URL below and confirm.
https://github.com/Encod3d-Sec/TORCH/tree/main/skills/hunt/hunt-ics
Enable the skill in any chat where you want it available.

Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills

git clone --depth 1 https://github.com/Encod3d-Sec/TORCH.git /tmp/TORCH
mkdir -p .claude/skills
cp -r /tmp/TORCH/skills/hunt/hunt-ics .claude/skills/hunt-ics

Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Hunt Ics in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Hunt Ics on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Hunt Ics is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

Loaded only when it is needed

The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

Switch models mid-chat

Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Hunt: ICS / SCADA / OT

Assumes hunt-core for the scope gate, two-account rule, confirmation gate, enumeration limits, stop conditions, wiki protocol, FIND output, and Deadends. Do not re-derive any of that here.

SAFETY - hard stop, overrides everything below

Writing to a live PLC moves real machinery: it can over-pressure a vessel, over-speed a motor, or disable a safety interlock, and so damage equipment or injure and kill people. Only ever drive a process toward a dangerous state on a target that is EXPLICITLY an authorized lab or CTF simulation. Never on a real plant, and never when scope carries no_dos/passive_only - there OT is READ-ONLY (FC1-4), never write (FC5/6/15/16). The instant the objective is met (e.g. the flag the HMI/CCTV reveals is in hand), STOP writing and revert any coil/register you changed if you can. This is a hard stop and it outranks every other note in this file.

Wiki


qmd_query "ICS SCADA OT Modbus S7comm EtherNet/IP DNP3 OpenPLC Node-RED PLC HMI coil holding register" via wiki-search MCP

Hub: [[network-moc]] (live index). Primary page: [[ics-scada-modbus]]. Payload arsenal: wiki/payloads/modbus.md. Anchors: [[iot-attacks]], [[firmware-hardware]].

Attack surface (ranked by exploitation value)

Modbus / TCP 502 - no auth, no session, trivially readable AND writable. The primary lever: FC1-4 read coils/registers, FC5/6/15/16 write them. Highest value, highest danger.
OpenPLC web (8080) and Node-RED (1880, dashboard /ui) - the SCADA/IT side that drives the PLC. OpenPLC routes /programs /monitoring /hardware /users; authed OpenPLC = upload-program RCE. Node-RED /ui socket.io config NAMES the registers.
HMI web app (Flask/Werkzeug with an /api/... live-state endpoint) - not itself the lever but the ORACLE that tells you whether a write landed.
S7comm / TCP 102 (S7/iso-tsap) - python-snap7 db_read for creds/flags in data blocks.
EtherNet/IP 44818, DNP3 20000, then BACnet 47808/udp, Fox 1911/4911, OPC-UA 4840 - enumerate with the matching nmap scripts; less commonly the CTF lever, but map them.

Methodology

Recon: nmap -p- ... then nmap -Pn -p102,502,20000,44818 --script s7-info,modbus-discover,enip-info,dnp3-info $T. Note the unit/slave id ("sid 0x1").
Find the oracle: the HMI /api/state-style endpoint (it reads the PLC and tells you live status). Node-RED /ui socket.io config NAMES the registers ("Read Pressure" = FC3 reg0). This is how you know a write landed.
Enumerate Modbus (raw socket, not pymodbus): dump FC1-4, ASCII-decode registers (creds/flags hide there), scan unit ids. Reads (FC1-4) are safe; this step never writes. See [[modbus]].
Map process variables -> registers/coils. The pressure/temp/level sensor is a holding/input register; pumps/valves/cooling/safety are coils.
Drive to the danger state (authorized lab/CTF ONLY - see SAFETY): FC6 over-drive the process variable to max (65535); then defeat the safety/protection interlock - a controller loop (cooling/relief/ESD) fights you, and it is usually ONE coil. Find it with a BOUNDED, one-write-at-a-time probe, NEVER a free-running write loop (hunt-core: never enumerate writes; this narrow, reverting, oracle-gated probe on an authorized simulation is the sole carve-out): write ONE coil (FC5), read the HMI oracle, and if the state did not flip toward the goal, write that coil back before trying the next. Cover coils 0-31 this way - the control coil is rarely at 0-5, so go wide but stay bounded. If nothing in 0-31 flips the interlock, that is a Deadend; do not widen indefinitely.
Read the payoff: HMI flips (e.g. "Explosion Detected!") and points the CCTV/HMI at new media. The flag is frequently a VISUAL overlay in that image/video - fetch it, ffmpeg -i x.mp4 -vf fps=1 f_%02d.jpg, and VIEW the frames (not strings/exif). Objective met -> stop writing (SAFETY).
S7: python-snap7 db_read the data blocks (creds/flags). Banner-only sims (s7-info works but no DBs) = decoy.
OpenPLC: try openplc:openplc; if changed, do NOT grind the login - the protection is usually disabled directly over Modbus (a coil flag the ladder reads). Authed OpenPLC = upload-program RCE.
Distill (when confirmed): stage a GENERIC candidate - python3 scripts/wiki-stage.py --kind technique --slug <slug> --target-page techniques/mobile-iot/ics-scada-modbus.md.

Chaining

HMI / Node-RED web app -> read the /api/state oracle and the register NAMES -> know exactly which Modbus address is the process variable and which is the interlock, instead of guessing blind.
OpenPLC authed (default openplc:openplc) -> upload-program RCE on the SCADA host -> shell, not just process control. Prefer disabling protection over unauthenticated Modbus first (cheaper, no login grind).
Danger state reached -> the payoff media (CCTV/HMI overlay) -> flag. The Modbus write is the pivot; the flag is in the frame (step 6).

Evasion

Wrong unit/slave id looks like a dead PLC: if FC1-4 error out, scan unit ids (the modbus-discover "sid" value) before concluding the device is read-only or absent.

Lessons (THM Kaboom)

The challenge NARRATIVE is the literal solve ("over-pressure the pump -> blow-out", "disable the protection"). Map each phrase to a Modbus write before chasing web creds.
Coils 0-5 were inert; the safety kill-switch was coil 10 -> sweep WIDE (still bounded and reverting, per step 5).
The OpenPLC login (openplc:openplc changed) was a rabbit hole that OOM'd the tooling host with a brute. Disable protection over Modbus instead.
Flag was burned into the CCTV explosion video (/video?mode=explodedflag23), invisible to strings -> extract + view frames.

Confirmation gate

NOT confirmation: a Modbus/S7/EtherNet-IP port open and banner-grabbed; a coil/register/data-block you can only READ; an FC1-4 dump that returns values; the HMI showing a value you did not change; a write (FC5/6/15/16) that returned an ACK with no observed state change. A readable register alone proves reachability, not control.

IS confirmation (control/impact finding): a coil/register you WROTE whose effect is observed on the oracle - the HMI / Node-RED live-state flips to the intended (danger) state, the simulated process variable moves, or the interlock is demonstrably defeated - reproduced on a clean connection, culminating in the payoff (the flag the HMI/CCTV reveals). No observed effect = no control finding; log it as reachability at most.

An unauthenticated READ of OT state is a valid lower-severity finding in its own right (confirmed by demonstrating the unauth read), but it is NOT a control or physical-impact claim - keep the two separate.

Severity

Demonstrated impact	Rating
Unauth write to a safety-critical coil/register - physical impact, ESD or interlock bypass	critical
Unauth process-variable write (no safety bypass)	high
Unauth read of OT state (registers/coils/data blocks)	medium

Deadends


Append: - [ ] ICS <host> -- <proto> read-only / sim-only; no writable control coil found (swept 0-31)

Record the range you swept and whether an oracle existed, so the next pass does not re-run the same bounded probe.

Frequently asked questions

What does the Hunt Ics AI skill do?

ICS/SCADA/OT exploitation - Modbus (502), S7comm (102), EtherNet/IP (44818), DNP3, OpenPLC, Node-RED SCADA, PLC/HMI/coil/holding-register attacks. Use when a target exposes industrial protocols or the goal is to drive a plant to a dangerous state (over-pressure/over-speed/disable interlock) and read the flag the HMI/CCTV reveals.

Why use Hunt Ics on TypingMind?

Because you install it once and use it with any model. Hunt Ics is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Hunt Ics in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/Encod3d-Sec/TORCH/tree/main/skills/hunt/hunt-ics. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Hunt Ics?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Hunt Ics?

As many as you like. As long as a model supports skills, you can use Hunt Ics with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Hunt Ics AI skill free?

Yes. It is published on GitHub by Encod3d-Sec under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

Overview

Installation

Install in TypingMind

Install in another agent

Use it in TypingMind

Works with any AI model

Skill instructions

Hunt: ICS / SCADA / OT

SAFETY - hard stop, overrides everything below

Wiki

Attack surface (ranked by exploitation value)

Methodology

Chaining

Evasion

Lessons (THM Kaboom)

Confirmation gate

Severity

Deadends

Frequently asked questions

Related AI Skills

Set up your own AI workspace now