thm-trainer
Skill2 starsChanged 4 months agoMITView on GitHub
The core decision loop for driving any TryHackMe room. Use at the START of every room, whenever the user asks "what next", and whenever progress stalls. Teaches classify -> observe -> hypothesize -> test -> decide, so the approach is chosen from what the room is actually teaching rather than defaulting to port-scan-then-privesc. This is the thinking layer above the methodology and command skills.
In this repository4 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- network-forensicsSkill
- thm-methodologySkill
- windows-ad-methodologySkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 5 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 16
TEST -> one focused action whose result decides the next move
- Distinctive: line 23
Read what the room tells you before touching a tool:
- Distinctive: line 57
No new info -> you're off the path.
- Distinctive: line 74
TEST: vhost fuzz — ffuf -H "Host: FUZZ.futurevera.thm" -fs <default>.
- Distinctive: line 88
Test — one focused action whose result decides the next move.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
THM Trainer — how to drive a room
A coaching guide, not a solutions list. It teaches the loop a good player runs on ANY room — familiar or not — so neither you nor Claude confidently does things that lead nowhere. The commands and agents are just tools this loop reaches for.
The loop
- CLASSIFY -> what is this room teaching? what's the goal (flag string vs shell)?
- OBSERVE -> the lightest thing that fits the class; don't over-scan
- HYPOTHESIZE-> "the way in is probably X, because Y" (one sentence, out loud)
- TEST -> one focused action whose result decides the next move
- DECIDE -> new info -> loop; goal met -> STOP; no new info -> reclassify
Underneath it all: every action should change what you do next. If a scan wouldn't change your plan, don't run it.
Beat 1 — CLASSIFY (the beat everyone skips)
Read what the room tells you before touching a tool:
- Room card / tags — "web", "subdomain enumeration", "Active Directory", "OSINT", "steg". The strongest signal. Read it first, every time.
- Title — "Takeover" -> subdomain/DNS takeover. Names hint.
- Task text — a domain name means DNS/vhost work; a downloadable file means offline analysis.
- a .pcap / capture file -> network-forensics skill + /pcap (tshark)
- Difficulty — "easy" usually means one technique, one insight. Don't over-engineer. Then name the GOAL: flag string (many recon/web/OSINT rooms — no shell needed) vs user+root shell (full compromise) vs a specific answer. Write it in notes.md before scanning.
For Claude: state the classification and goal in your first message, before any tool. If the category isn't given, infer it from the card or ask — don't default to a port scan.
Beat 2 — OBSERVE (enough, not everything)
Match the observation to the class. A port scan is the default for a service box, not always the right first move. Signal -> first observation:
- domain name (foo.thm) -> /etc/hosts, vhost fuzz, cert SANs (/web-recon)
- 80/443 + a real website -> browse it, view source, dirs, params (/enum-web)
- 88 + 389 + 445 -> Active Directory (/enum-ad)
- 22 + a versioned service -> version -> CVE lookup
- a downloadable file -> file/strings/exiftool/binwalk (/steg)
- a .pcap / capture file -> network-forensics skill + /pcap (tshark)
- "find the answer to..." -> OSINT / careful reading
Beat 3 — HYPOTHESIZE (say it before you test)
One sentence: "the way in is probably X, because Y." If you can't finish it, observe more.
For Claude: state the hypothesis before the tool call, not after.
Beat 4 — TEST (one focused action)
The single action that best confirms or kills the hypothesis. Not three tools in parallel — that's usually avoidance. The test should read clearly yes/no.
Beat 5 — DECIDE (and know when to stop)
- New info -> loop back to Beat 2 with a new hypothesis.
- Goal met -> STOP. Got the flag the room asked for? Done. Don't escalate to a shell you don't need.
- No new info -> you're off the path.
The stuck signal
You're off the path when you repeat actions without new information — another wordlist, another tool, another port, learning nothing. That churn feels like progress and isn't. When you feel it, re-run Beat 1: did I classify right? am I chasing a shell when the goal was a flag? what is the room tagged, and am I doing that technique? did I skip a hint?
For Claude: if two consecutive actions produce no new information, stop and say so: "I'm not learning anything new — let me reclassify." Never fire tools to look busy.
Worked example: "Takeover"
- CLASSIFY: tagged subdomain enumeration; title "Takeover"; task about futurevera.thm. Goal: one flag, no shell. -> we will not ssh into anything.
- OBSERVE: signal is a domain, so first move is resolving it — add futurevera.thm to /etc/hosts, browse. Main site is a dead end, which fits a recon room.
- HYPOTHESIZE: "probably a forgotten subdomain, because the room is about subdomain takeover and the main site is a dead end."
- TEST: vhost fuzz — ffuf -H "Host: FUZZ.futurevera.thm" -fs
<default>. support.futurevera.thm appears. New info -> loop. - DECIDE -> loop: add it to hosts, observe. The artifact is the SSL cert — read its SANs, which reveal another subdomain. Visit it -> redirects to an AWS S3 URL with the flag in the URL. Goal met -> STOP. Elapsed with the loop: minutes. The tools (ffuf, openssl) were always in the kit. What was missing was Beat 1 — classifying as recon and naming the goal as a flag, which stops you sshing into a box that was never about shells. The lesson generalizes: classify before acting, hypothesize before testing, stop when the goal is met.
One-screen version
- Classify — read the card. What's it teaching? Goal (flag vs shell)? Write it down.
- Observe — lightest thing that fits the class. Don't over-scan.
- Hypothesize — "the way in is probably X because Y." One sentence.
- Test — one focused action whose result decides the next move.
- Decide — new info -> loop; goal met -> stop; no new info -> reclassify. Stuck = repeating actions with no new information. Then go back to step 1. The flag is the finish line — know which flag you're chasing.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

