Zhangir Ospanov

I am a cybersecurity researcher, with a passion for Red Team, MalDev & Threat Hunting/Threat Intelligence.

[Zhangir Ospanov @s0ld133rr]

@MAXdeg0 Thanks for sharing! Would be a pleasure for me to get feedback on the repo issues/dm 🤝

[Ian Bell @ibell63]

@MSNightmare2000

[Zhangir Ospanov @s0ld133rr]

Recently, @Tailscale released tailcat - a netcat style tool working over TailScale network

I made a tiny wrapper to make stable SOCKS5 reverse proxy over tailcat, it’s useful when you need to pivot inside the network during pentest/red team

Link: https://t.co/PC3lYHRTLv

[Zhangir Ospanov @s0ld133rr]

@bitcoin180 Hi! Thanks for your feedback, I will find out the root cause and try to fix it

[Zhangir Ospanov @s0ld133rr]

@AelsMartin Thanks, suggestions are always welcome 🫂

[Zhangir Ospanov @s0ld133rr]

@AelsMartin @_aircorridor @co11ateral @three_cube @DI0256 @IamSmouk Hi, thanks for your feedback I will try to fix it 😅👀

[Zhangir Ospanov @s0ld133rr]

@support_huihui Try to use PentestCode

[Zhangir Ospanov @s0ld133rr]

@ZepingLiu79973 @ClaudeDevs @claudeai It’s annoying tbh

[Zhangir Ospanov @s0ld133rr]

Why whenever I try to apply for CVP I got auto-reject from Claude? I don’t even know why 😵

@ClaudeDevs @claudeai it’s unclear for me, really

[Zhangir Ospanov @s0ld133rr]

@ifipls Sure! I can share it with you, I will text you later

[Zhangir Ospanov @s0ld133rr]

While reviewing forks of my PentestCode repository, I found a malicious fork containing a multi-stage JavaScript backdoor.

⚠️ The original PentestCode repository was NOT compromised. The malware existed only in a third-party fork.

- Dynamic C2 resolution from Ethereum transactions
- Multi-stage execution chain (Stage 1 → Stage 2 → Stage 3)
- Arbitrary JavaScript execution via eval() and spawn()
- WSL2-to-Windows host pivot using node.exe
- Stealthy detached background execution

The infrastructure appears to target pentesting and AI tools, including forks of PentestCode, KaliGPT, Kittysploit, Meterless, Nebula, Codex and others.

IoCs:

Ethereum
0x33ff3edaf55a8e03dcbc7cb40d498a49cd499891

Current C2

23.27.20.187:443
23.27.20.187:80

SHA256 

950a8f79e48cb6b36d731bb7adf695e2420014813d5705ce41ea8090e8b06724
62b19e0cbb110bc5d33ad4db07af9c7d1018c3951350369d6bdf2b2c1ebc2828
d15c51eaeecc0230313565dd4457f1dfcc57cfac007d519cd1789cfc2af599eb

Looks like an interesting GitHub supply-chain campaign. Full technical write-up and additional IoCs coming soon.

[Zhangir Ospanov @s0ld133rr]

@ifipls Hi, currently I don’t have updates cause switched for another more interesting research

[Zhangir Ospanov @s0ld133rr]

@_jensec @AnthropicAI @cjav_dev @mikeyk CVP is a random, cause 7 times I tried to pass CVP and had auto reject 🥲

[Zhangir Ospanov @s0ld133rr]

@MichalKoczwara They targeted AI & Pentest tools for delivery

[Zhangir Ospanov @s0ld133rr]

@cocomelonckz @defcon GOAT

[Zhangir Ospanov @s0ld133rr]

PentestCode v0.2.4 released

This release focuses on tightening the multi-agent architecture and making agent boundaries explicit instead of relying on prompt behavior.

Changes:

- Added an Artifact Store. PoCs generated by one subagent are now registered in the engagement state and reused by other agents instead of being regenerated.
- Reworked engagement state delivery. Subagents now receive only task-relevant context, reducing token usage.
- The coordinator now acts strictly as a coordinator. Added guardrails, watchdogs, execution limits, and loop detection.
- Strengthened result validation. Claiming “root” is no longer enough, successful actions must be backed by artifacts, otherwise they’re marked as suspected.
- Refined reconnaissance skills and adjusted the loading order.

Update:
pentestcode upgrade

Feedback, bug reports, and interesting success/failure cases are always welcome.

https://t.co/x7pOz3WOAb

#CyberSecurity #Pentesting #LLM #OpenSource

[Zhangir Ospanov @s0ld133rr]

@co11ateral Hi man, thanks! I am preparing v0.2.4 with major updates 👻

[Zhangir Ospanov @s0ld133rr]

PentestCode v0.2.3 👾

Better exploitation methodology, significantly lower context usage, and more reliable reasoning.

Highlights:

• Route-first exploitation instead of request-shape guessing
• Version-diff reconnaissance
• Exact version fingerprinting before source analysis
• Attempt memory per exploitation vector
• Universal tool-output caps
• Cached static prompts
• Reference-based tool output
• Typed subagent result trailers
• Evidence-based confidence scoring
• Deterministic regression tests

Benchmarks:

-69% scan context
-99% resolved-vectors payload
-94% context growth

https://t.co/x7pOz3WOAb

[Zhangir Ospanov @s0ld133rr]

🧵 PentestCode v0.2.2 — a token-efficiency release. The AI pentest agent got leaner without getting dumber.

⚙️ The deterministic orchestrator is now ON by default. The coordinator dispatches subagent waves through a rolling pipeline — a slot refills the instant one finishes, so you get bounded concurrency with full depth.

Real multi-host pivot run: -87% provider overloads.

💸 Cheap-model offload. Big raw tool dumps (nmap/nuclei/gobuster) get digested by a small model before they hit the transcript — the raw output stays retrievable by ref, and reasoning always runs on your main model.

Turn it on with /small-model (or the small_model config key).

🧠 Smarter, cheaper loops:
• resolved-vectors ledger — the agent stops re-testing dead ends
• per-provider concurrency cap — no more 429/529 storms on fan-out
• objective-based task prompts + cached system prefix = smaller context every turn

🎛 You stay in control:
• abandon a subagent (let it finish) vs kill (hard-stop, reclaim tokens)
• orchestrator off-switch: OPENCODE_DISABLE_ORCHESTRATOR=true
• tune fan-out: OPENCODE_ORCHESTRATOR_CONCURRENCY (default 3)

https://t.co/kmxbIKt4dF

And yes, now it’s in beta stage - not alpha anymore 👾