Pular para o conteúdo
Novo

DeepSeek V4.1 Flash grátis por 7 dias: 10M tokens por dia após recarga de $10 ou mais. Até 30/11.

Como funciona
aeondave/technique-ctfEngenharia reversa e fuzzing

technique-ctf

⁨Lab/CTF: general challenge-solving methodology across every category; artifact triage, spec/protocol identification, oracle-driven iteration from target error messages, minimal-matrix brute forcing, harness building with pwntools, parallelizing independent challenges/variants with subagents, and saf⁩

v2026.11.13-21cd5683 instalações+3 em 7 dias39 EstrelasMITsincronizado há 10 horas

Instalar

Ainda não tem a CLI da NotHoly? Instale primeiro.

$ curl -fsSL https://notholy.com/install.sh | sh
$ notholy skills install aeondave/technique-ctf

Grava em ⁦~/.notholy/skills⁩ depois de conferir a assinatura e o SHA-256.

AssinaturaAssinado, ed25519
KW10kwzwvv…fVMtrJBw==A CLI confere o SHA-256 e a assinatura antes de gravar qualquer arquivo.
Pacote7b89a34c…f5a645e9
7b89a34c6607…d5aaf5a645e92 arquivos · ⁦6.0 KB⁩
VarreduraMédio1 achado
  • injection.no_confirmationSKILL.md:95
Ver a varredura completa
Fontegithub.com/AeonDave/malskill/tree/main/offensive-ctf/technique-ctf@21cd568

Este item ainda não foi escrito no seu idioma, então aparece em English.

Solve any flag-style challenge with a category-agnostic loop: triage the artifact, identify the exact spec or format, drive the smallest correct interaction, and let the target's own responses steer each next input — then verify and submit the flag.

When this skill applies

  • Any controlled lab/CTF objective: a remote service, a downloadable bundle, a puzzle artifact, or a skeleton client to complete.
  • Before reaching for a domain skill, to decide which *-ctf skill(s) to load and how to structure the attack.
  • Whenever a target emits parseable feedback — validation errors, diffs, status codes, timing, or partial output — that can be used as an oracle.
  • When several independent challenges (or independent parameter variants of one challenge) can be worked in parallel.

Operating model

  1. Triage. Inventory what you were given: hash and list files, read the brief literally, note the target host:port, the stated flag format, and every concrete value (IDs, keys, payload strings, ports). Identify the dominant artifact class — it selects the domain skill.
  2. Identify the spec. Name the exact format/protocol/algorithm before writing an exploit. Match magic bytes, headers, wording, and cited standards. The brief usually states the spec and the exact expected input — build to that, not to placeholder values in any example code.
  3. Minimal correct interaction. Send the smallest well-formed unit and observe. Probe liveness both ways: some services greet with a banner, others reply only after you send — send-then-recv, do not hang waiting for a banner.
  4. Read the oracle. Treat every response as signal. Precise validators say exactly what is wrong (expected X but got Y, invalid state, a stack trace, a length, a 401 vs 403, a timing delta). Walk the error chain field by field; each rejection names the next thing to fix. This beats guessing and beats blind brute force.
  5. Iterate / brute the smallest unknown. When one field is genuinely ambiguous after the oracle is exhausted, brute only that minimal space (e.g. a counter × a type bit), reconnecting/resetting state per attempt. Never brute a large space blindly when a signal exists.
  6. Extract and verify. Pull the flag with a format regex, sanity-check it, and only then submit. If a submission API exists, use it; confirm acceptance. Do not paste flags into external services.

Cognitive stance (negative constraints)

  • Evidence first. Do not claim a solve, a working exploit, or a root cause without a reproduced observation. If a step was skipped or failed, say so.
  • Oracle over guessing. Exhaust the target's feedback before brute force; exhaust a small brute before a large one.
  • Minimal action. Prefer offline decode of a supplied artifact over live interaction; prefer one crafted request over a scanner. Loud, broad tooling wastes time and corrupts stateful targets.
  • No rabbit holes. Time-box a dead end. If the oracle stops yielding, re-triage: wrong layer, wrong spec, or an assumption to test — not more of the same input.
  • Verify the flag. A plausible-looking string is not a flag until it matches the stated format and (if possible) is accepted.

Category routing

Load the domain skill that matches the dominant artifact; load more than one only for genuinely cross-domain bundles.

Cue Load
Web app, HTTP, cookies, SSRF/SQLi/XSS web-ctf
Ciphers, RSA/ECC, hashes, custom crypto crypto-ctf
Binary exploitation, heap, shellcode, ROP pwn-ctf
Disassembly, unpacking, license/keygen reverse-ctf
Disk/memory images, PCAP, stego, carving forensics-ctf
Spacecraft, CCSDS, telemetry/telecommand, TLE satellite-ctf
Modbus/S7/BACnet, PLC, SCADA, OT protocols ics-ctf
Logic traces, UART/SPI/JTAG, firmware, RF/SDR hardware-ctf
APK/IPA, mobile traffic, Frida mobile-ctf
Smart contracts, EVM, on-chain blockchain-ctf
Prompt injection, model files, embeddings ai-ml-ctf
Malware sample triage in a lab malware-ctf
Unity/IL2CPP, native game binaries, saves, GamePwn game-ctf
Subdomains, people, leaked creds, public sources osint-ctf
Jails (bash/py), encodings, DNS, misc puzzles misc-ctf
Cloud/IAM, metadata, buckets cloud-ctf
Writing up the solve writeup-ctf

Parallelization

Independent work should run concurrently, not serially.

  • Dispatch one subagent per independent challenge, or per independent variant of a stubborn challenge (different spec guess, different parameter branch). Give each the concrete target, the exact spec, and any reusable code so it does not rediscover the basics.
  • Split a single challenge only along a true seam (recon vs. exploit dev, or independent sub-flags) with no shared mutable state.
  • Keep results, not transcripts: each subagent returns the flag and the minimal working method; you retain the conclusion.

Tool routing

  • Remote loop: pwntools (remote, recvuntil, sendline), or raw socket for exact byte control. Script every interaction — never solve a stateful service by hand.
  • Transforms: CyberChef, xxd/hexdump, base64/base32, jq, Python struct/construct for framed formats, binwalk/file/strings for unknown blobs.
  • Recon of the artifact, not the network: read the files first. Reserve scanners for when the brief points at a broad service surface, and keep them scoped.
  • Reproducibility: keep a single script per challenge that goes from connect to flag, so a fix is one edit and a re-run.

Flag discipline

  • Match the stated format; default regexes: HTB\{[^}]+\}, flag\{[^}]+\}, [A-Za-z0-9_]+\{[^}]+\}, or the exact prefix the platform uses.
  • Verify before submitting; if the platform has a submit API/tool, use it and confirm the acceptance message.
  • Do not exfiltrate flags, source, or target data to external/third-party services; keep everything in the lab context.

Quick pivots

  • Encrypted challenge bundle (zip/7z): check the challenge brief, download page and platform instructions for the supplied archive password before attempting recovery. Do not infer a password from an account email, team name or a downloader's default.
  • Timeout on connect: the service wants input first — send the minimal frame/request, then read.
  • "Invalid" on a unit you believe is valid: an encoding assumption is wrong (raw byte vs. ASCII string, endianness, off-by-one length, CRC seed). Try the alternate encoding before rebuilding everything.
  • Errors reference a counter/sequence/state: the target is a state machine — advance every counter it tracks in lockstep; reconnect to reset when you lose sync.
  • Same error regardless of input: you are at the wrong layer or failing an earlier gate (auth, framing, checksum) before your payload is even parsed.
  • Big search space: find the oracle that collapses it (an error that leaks length/position, a timing side channel, a partial match) instead of brute forcing.
  • Stuck: re-read the brief for an unused concrete value; it is almost always load-bearing.
  • Total silence is a different failure from a rejection. A target that never answers may be broken or not listening, not wrong-input. Before iterating on payloads, prove the far side is alive and consuming: send a deliberately malformed unit and check you still get some reaction, verify the peer actually drains what you send (a queue that accepts data proves buffering, not processing), and plant a canary that needs the minimum machinery to echo. Endless silence with zero diagnostics usually means the environment, not the exploit.
  • Verify the instance is yours before trusting it. A platform that allows only one live instance at a time will silently replace or expire the previous one while its status still reads "ready", so a cached ip:port can point at a dead or foreign container. Reconnect and match the banner before concluding a technique failed, and serialize work when instances are exclusive.
  • Re-derive a parameter per instance when the target regenerates it. If a value changes between spawns (key, IV, nonce, TLE, port), any artifact captured from an earlier instance is poison — collect every dependent input from the same live instance.

Resources

  • references/harness-and-oracle-patterns.md — reusable pwntools/socket remote-loop templates, an oracle-error-walking pattern, a minimal-matrix brute template that reconnects per attempt, flag-regex helpers, and a subagent parallelization pattern for independent challenges. Load when building the solve script for a remote or stateful challenge.

Arquivos desta versão

Permissões

PermissãoDeclaradaEncontrada na varredura
Lê arquivosNãoNão
Grava arquivosNãoNão
Roda comandos no shellNãoNão
Usa a redeNãoNão
Usa conectoresNãoNão
Controla um navegadorNãoNão

Achados da varredura

Revisado em 13 de nov. de 2026

  1. MédiaPrompt injection · injection.no_confirmationSKILL.md:95

Uma varredura limpa reduz o risco, mas não prova que o pacote é seguro. Leia o que você instala.

VersãoPublicada emCommitSHA-256AssinaturaMudança de permissão
v2026.11.13-21cd56813 de nov. de 202621cd5687b89a34cAssinada–

Pergunte tudo o que o seu escopo permite.

Crie a conta, adicione créditos e faça a primeira pergunta no chat ou no terminal.

  1. 1/3

    Crie a conta

    Individual ou empresa. Dá para virar empresa depois sem perder saldo nem chaves.

    Pessoa físicaEmpresa
  2. 2/3

    Adicione créditos

    No cartão. Você paga por token, sem assinatura.

    $5.00recarga mínima
  3. 3/3

    Pergunte

    No chat, ou instale o CLI e rode notholy login no terminal.

    $ notholy login