Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add pashki975/thm-claude-kit --skill thm-trainergit clone --depth 1 https://github.com/pashki975/thm-claude-kitWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/pashki975/thm-claude-kit/thm-trainer)<a href="https://agentmods.dev/skills/pashki975/thm-claude-kit/thm-trainer"><img src="https://agentmods.dev/badge/skills/pashki975/thm-claude-kit/thm-trainer/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/pashki975/thm-claude-kit/thm-trainer"><img src="https://agentmods.dev/badge/skills/pashki975/thm-claude-kit/thm-trainer.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00089 | $0.01429 |
| Opus 5 | $0.00044 | $0.00714 |
| Sonnet 5 | $0.00018 | $0.00286 |
| Haiku 4.5 | $0.00009 | $0.00143 |
Grade A, and why
thm-trainer scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 8d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 92 lines — stays where its author put it; the contents beside it link to each section on GitHub.
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
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 8d ago First seen · 92 lines · 89 tokens per session scan A 5d3f32e1a55b
thm-trainer is a skill published in the GitHub repository pashki975/thm-claude-kit (2 stars, last pushed 25d ago), licensed MIT. It adds 89 tokens to every session and 1,429 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
offensive-vuln-classes
Exploit development curriculum covering core vulnerability classes with real-world CVE case studies: stack/heap buffer overflows, use-after-free, integer overflows, format strings, type confusion, and race conditions. Use when learning or teaching vuln classes, researching specific CVE patterns, or building exploit…
stuck
Diagnoses a broken course setup by reading the folder, so the student never has to describe the problem in English. Takes no argument. Reports which class they are on and which files are filled, then names one blocker and one fix: Claude Code opened at the wrong level, .claude/skills/ missing, skills not loading…
learning-loop
(Industry standard: Loop Agent / Single Agent) Primary Use Case: Self-contained research, content generation, and exploration where no inner delegation is required. Self-directed research and knowledge capture loop. Use when: starting a session (Orientation), performing research (Synthesis), or closing a session…
eat
Extract knowledge, frameworks, and methodologies from a URL, file, video, article, or podcast. Use when user says '/eat', 'eat this', 'eat from', shares a URL/file for insights, or wants to learn from a video/article without reading the whole thing. NOT for summarization or news digests. Requires yt-dlp, whisper or…
canvas-homework
Check Canvas for outstanding assignments and either surface the assignment, or draft a response in Max's voice. Use whenever Max asks what assignments / homework / coursework he still has to do, asks to pull an assignment's instructions or files from Canvas, or asks to start/draft an assignment. Orchestrates the…
myopenmath
Work a MyOpenMath / IMathAS math homework set that is launched from Canvas — read the questions, compute the answers, fill the boxes, and submit. Use whenever Max points at a MyOpenMath assignment, says a math homework is "in my Canvas", asks to do/finish/check a numbered homework like "2.1" or "2.2/2.3", or when a…