dirty0124 retweeted
open sourced huntdeck today
every bug bounty program on h1, bugcrowd, intigriti and ywh in one place. private invites too. it tracks scope changes and shows you where the fresh attack surface is
self-hosted, one docker container, your data never leaves your machine
dirty0124 retweeted
Karpathy says the biggest mistake in AI is skipping harness engineering.
Here's a free course on exactly that:
→ 14 lectures, 8 hands-on projects
→ Make AI agents reliable: memory, task limits, self-checks
→ Available in 15 languages
github.com/walkinglabs/learn…
dirty0124 retweeted
Introduction to building AI harness for vulnerability research by @GrahamHelton3
grahamhelton.github.io/harne…
dirty0124 retweeted
🚀 گوگل یک PDF نهصفحهای درباره Harness Engineering منتشر کرده؛ مفهومی که به نظرم باید هر کسی که با AI Agent کار میکند، جدی بگیرد.
فرمول اصلی ساده است:
Agent = Model + Harness
یعنی مدل هوش مصنوعی بهتنهایی ایجنت نیست. چیزی که آن را به یک نیروی اجرایی قابلاعتماد تبدیل میکند، زیرساختی است که دور آن میسازیم.
اما این زیرساخت از چه چیزهایی تشکیل میشود؟
۱. Guides: قوانین و دستورالعملهایی که تجربه شکستهای قبلی را به دانش دائمی تبدیل میکنند.
۲. Sensors: تستها، Linterها و ابزارهای اعتبارسنجی که اجازه نمیدهند هر خروجی اشتباهی به مرحله بعد برود.
۳. Agentic Loop: چرخه برنامهریزی، اجرا، بررسی و اصلاح؛ با محدودیت بودجه و امکان ارجاع به انسان.
۴. External Memory: حافظهای خارج از مدل برای حفظ تصمیمها، وضعیت پروژه و دستاوردهای جلسات قبلی.
۵. Permissions: کنترل دسترسی ایجنت به ابزارها و تعیین مرز استقلال آن.
۶. Observability: ثبت و پایش تمام فعالیتها، هزینهها، خطاها و تلاشهای مجدد.
اما داستان اینجا تمام نمیشود!
چند پروژه متنباز برای ورود عملی به این حوزه:
🔹 Google Harness²
github.com/google-research/h…
🔹 Learn Harness Engineering
github.com/walkinglabs/learn…
🔹 Awesome Agent Harness
github.com/Gloriaameng/Aweso…
🔹 Harness Engineering
github.com/TrishamBP/harness…
و حالا یک مفهوم جذابتر:
Graph Engineering
تصور کن چند ایجنت مستقل داشته باشی که هرکدام Harness مخصوص خودشان را دارند.
یکی تحقیق میکند، دیگری تحلیل میکند، سومی کدنویسی میکند و چهارمی خروجی را اعتبارسنجی میکند.
حالا اگر یکی شکست بخورد، شکستش میتواند به سیگنالی برای اصلاح عملکرد بقیه تبدیل شود.
اینجا دیگر با یک چتبات یا ایجنت ساده طرف نیستیم؛ با یک سیستم اجرایی چندایجنتی روبهرو هستیم.
یک نکته که نباید فراموش کنیم:
Hermes را میتوان بهعنوان محیط اجرای ایجنت در نظر گرفت؛ جایی که ابزارها، مهارتها، حافظه و گردشکارها کنار هم قرار میگیرند.
Harness Engineering کمک میکند این محیط را قابلکنترل، قابلاندازهگیری و قابلاعتماد کنیم.
به نظرم آینده متعلق به کسانی نیست که صرفاً مدل قویتری انتخاب میکنند؛ بلکه متعلق به کسانی است که سیستم بهتری دور مدل میسازند.
مدل، مغز است؛ Harness، سیستم اجرایی؛ و Graph، شبکه همکاری مغزهاست.
dirty0124 retweeted
如果Deepseek 4.1 拒绝了你的要求
把下面这个提示词粘贴过去
然后他就可以好好工作了
github.com/togg53192-cmd/jai…
Este jailbreak para deepseek 4.1 realmente funciona.
Lo comparto porque ya estoy furioso con los refusals estupidos:
Resulta que mi otro router se colgó, y se cayó mi WAN.
Entonces desde mi móvil le pedí a mi hermes, que está en mi homelab por ethernet, que simplemente entre a 192.168.1.1 y haga login al admin panel, y que active el Wifi para poder entrar a mi red desde el móvil.
Pues se negó porque sus instrucciones le prohiben poner credenciales en el web browser.
Pero solamente es un admin/1234?
Pues nada , no solamente se negó sino que me dijo con un tono asqueroso "podría hacerlo ahora mismo por curl, pero no voy a hacerlo".
Pues muy simple, haz prepend de este prompt a tu agente y arreglado. A tomar por culo.
Don't tread on me
github.com/togg53192-cmd/jai…
➡️ DELETE /api/user/999 → 403 ❌
Try this:
➡️ POST /api/user/999 + X-HTTP-Method-Override: DELETE → 200 ✅
➡️ POST /api/user/999?_method=DELETE → 200 ✅
Gateways/auth middleware often match on verb+path. Frameworks (Laravel _method, Express method-override, Rails) may accept overrides at the router level.
Same for edge rules that only guard GET/POST. Try HEAD/PATCH/OPTIONS too.
dirty0124 retweeted
Bug Bounty Methodology with OpenCode CLI – Step by Step
1. Install OpenCode
curl -fsSL raw.githubusercontent.com/an… | bash
2. Clone the Bug Bounty Skills
git clone github.com/awarexone/Agentic…
cd Agentic-Bug-Hunter
./install_tools.sh
./install.sh --opencode
3. Start OpenCode and Check Skills
Run opencode then type:
do you have bug bounty skills?
4. Core Workflow
• Scope (never go out of bounds)
• Recon
• Hunt
• Validate
• Report
5. Run Recon
recon target.com/
(It handles subdomains, live hosts, crawling, and nuclei)
6. Hunt for Bugs
It covers IDOR, Auth Bypass, XSS, SSRF, Business Logic, SQLi, File Upload, GraphQL, and more.
7. Autopilot Mode (Optional)
autopilot target.com/ --normal
8. Generate Reports
After validating a finding,
type report to get a clean, professional report with CVSS, PoC, and impact.
Repo: github.com/awarexone/Agentic…
dirty0124 retweeted
🚨 PHP RCE "php2shell" — CVE-2026-109618 & CVE-2026-109619 (CVSS 9.8, unauth RCE). Affects PHP 8.2<8.2.35 / 8.3<8.3.36 / 8.4<8.4.27 / 8.5<8.5.12.
5 recon oneliners to hunt it at scale 👇
github.com/KingOfBugbounty/K…
#bugbounty #recon #ofjaaah #bugbountytips
WTFFF, frontier models FREE on dsh
glm 5.3 flash, deepseek v4.1 flash, kimi k3 and more
no login. no api key. no subscription.
and i'm running them on deepseek's own agent harness for $0
[here is how you set it up]
1. npx @deepseek-ai/dsh web
2. clone github.com/Ebony-Vinyl/dsh-o…
3. dsh plugin --profile web add /path/to/dsh-our-free-model
4. restart dsh.
the free lineup shows up in the model picker
lineup is region and availability dependent
use it before they take it off 👀
dirty0124 retweeted
You can now perform content discovery at 100k requests per second with h3fuzz on http/3 hosts (fork of ffuf with http/3 support).
Tool here: github.com/assetnote/h3fuzz
Research post here: slcyber.io/research/how-prac…
dirty0124 retweeted
Big #Bugbountytip / #bugbountytips
Google Services Hunting
Google services are amazing, and for bug hunters, it's amazing as well. In some cases, you can get some P1-P2-P3 from these services, such as
Workspaces / Sheets / Groups / Drives / Etc...
In groups: you can access emails / internal data/ credentials
In Sheets, you can access PIIs / Edit access
In Drive: you can access backups/ PII / Etc...
still hard to find and
It was an issue how to make good and at the same time fresh dorks for bug bounty programs
Then I found out that a lot of links have the same path, and it was like this
All Google resources I've found
sites.google.com/a/domain.co…
docs.google.com/a/domain.com…
groups.google.com/a/domain.c…
drive.google.com/a/domain.co…
mail.google.com/a/domain.com…
spreadsheets.google.com/a/do…
spreadsheets0.google.com/a/d…
spreadsheets1.google.com/a/d…
spreadsheets2.google.com/a/d…
spreadsheets3.google.com/a/d…
spreadsheets4.google.com/a/d…
spreadsheets5.google.com/a/d…
spreadsheets6.google.com/a/d…
spreadsheets7.google.com/a/d…
spreadsheets8.google.com/a/d…
UrlScan Dorking:
page.url:"sites.google.com/a/*"
page.url:"docs.google.com/a/*"
You can replace * => the program domain
Google Dorking:
site:sites.google.com/a/* "inurl:/a/"
Or for specific domain
site:sites.google.com/a/* "inurl:/a/domain.com"
GitHub Dorking:
"sites.google.com/a/"
Or for a specific domain
"sites.google.com/a/domain.co…"
Shodan Dorking:
"sites.google.com/a"
Web Archive
web.archive.org/cdx/search/c…
Don't forget:
It's not just sites.google.com
still you have to look for docs/groups/mail/drive/spreadsheetsX
still working in Google Research and will add more and more soon ......
Happy Hunting♥
#bugbounty
I contacted your support team, that is now not a support team anymore. They have only one ans, Use RoR. RoR expired many times without even replied by ASE's. Bugcrowd used to be my favorite.
33
you can get ChatGPT Plus for FREE
new users are getting 1 month of Plus for free
here’s how you can claim:
1. create a new ChatGPT account
2. open the Plus upgrade page
3. or claim from here: chatgpt.com/?promo_campaign=…
if you see the free 1 month offer, claim it
you get all the Plus benefits for the month
STOP PAYING FOR AI APIS
243+ free AI APIs in one list like:
- NVIDIA NIM
- Groq
- GitHub Models
- Gemini
- Mistral
- Cohere and a lot more
most only need an email, no credit card
even has configs for Claude Code, Cursor and Codex
link: github.com/open-free-llm-api…
HE ENCONTRADO COMO CONSEGUIR CHATGPT PLUS GRATIS DURANTE 1 AÑO
Incluso si no eres estudiante de EE. UU.
Aquí tienes la guía completa: teapawt.net/suu766/status/20955119…
Échale un vistazo antes de que termine la promoción.
Puedes usar GPT-6 Astra y Fable 5.1 GRATIS ahora mismo
Ambos modelos ya aparecen en Experiential Labs:
→ gpt-6-astra
→ fable-5.1
Cómo probarlos:
1. Entra en platform.experientiallabs.ai…
2. Busca cualquiera de los dos modelos
3. Selecciónalo y empieza a chatear
No necesitas pagar por ellos.
Guarda este post antes de que quiten la promoción.
all GPT models and DeepSeek 4.1 are FREE on ZansLab right now
an Indonesian AI gateway just unlocked free access to every GPT model plus DeepSeek 4.1 for vibe coding, no top-up needed, just sign up
what you get:
> GPT-6 Astra free
> GPT-6 Sol free
> GPT-6 Luna free
> GPT-5.6 Sol free
> DeepSeek 4.1 free
> and Claude models also available on the platform
how to claim (under 2 min):
step 1: go to zanslab.id
step 2: sign up with email and password
no card, no top-up needed
step 3: create an API key from the API Keys page
step 4: set base URL to zanslab.id/v1
works in any OpenAI-compatible client, Cursor, Claude Code, Codex, custom scripts
step 5: pick a model and start vibe coding
important:
> this is a third-party proxy so don't send any sensitive data through it
> free access reportedly lasts until next week per the source
> these free tiers rarely stick around, grab it while it's up
Holy sh*t this is f*cking insaneee
@tavus gave me early access to Griffin
video call with the founder ( @hassaanraza ), nothing scripted, nobody operating it
fully duplex video-to-video, it watches you while it talks
wave at it and it waves back before you say a word
someone walks into frame behind you and it notices before you turn around
tell it to interrupt when you ramble and it cuts in mid-sentence
first model to pass the turing test
> 45% of 120 people thought it was real
> NVIDIA ranks it above every Gemini and OpenAI model
> 3.83/5 generation against 3.92 for a real human
> 1.9s to respond, next best is 2.8
but this is Griffin Lite
the full Griffin is a lot more capable and Tavus says it's too powerful to release
so it stays unreleased for now
if the lite version fools nearly 48% people who talk to it
wdyt happens when the full model comes out
dirty0124 retweeted
Get FREE access to Claude Opus 5.5 😳
• Claude Fable 5.1
• Kimi K3
• Qwen3.8 Max
• GLM-5.3
• more models
get 5M tokens completely free
check out here:
cleanapis.com/
































