Security audit

CTF Web & Reverse Audit

Security checks for vulnerabilities and agentic risk

Overview

This skill is a static CTF security reference, but it includes broad, actionable exploit and bypass playbooks without enough scoping or safety warnings.

Install only if you want an offensive CTF/security-research reference and will use it strictly on owned, lab, or explicitly authorized targets. Review the broad reverse-engineering, malware, automotive, and third-party upload sections before use; prefer isolated VMs/containers and avoid applying the commands to real systems or sensitive binaries.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands

Findings (105)

Description-Behavior Mismatch

Medium

Confidence: 89% confidence
Finding: The file substantially exceeds the declared scope of a web attack and reverse-audit helper by aggregating broad reverse-engineering, anti-analysis, malware-adjacent, and platform bypass tradecraft. Even if framed as CTF notes, this expands the agent's capability surface into dual-use areas that can assist real-world evasion or unauthorized software tampering.

Context-Inappropriate Capability

High

Confidence: 97% confidence
Finding: These lines provide explicit anti-analysis and malware/C2-relevant guidance, including bypass and capture techniques that can be repurposed against real malware samples or production software. Because the skill is not narrowly scoped to defensive sandboxing or authorized incident response, the content increases the risk of misuse beyond the advertised web-audit role.

Context-Inappropriate Capability

Medium

Confidence: 95% confidence
Finding: This section goes beyond passive reverse-engineering and describes an operational workflow for patching a Go C2 client, registering to a server, enumerating other clients, and downloading their files. That is actionable post-compromise tradecraft enabling unauthorized access and data collection, not merely analysis guidance.

Context-Inappropriate Capability

Medium

Confidence: 93% confidence
Finding: The content teaches a concrete seccomp bypass by overriding libc symbols with Rust `#[no_mangle]` to disable sandboxing. This is a sandbox-escape technique that can be repurposed outside CTFs and exceeds the stated web/reverse audit purpose by enabling evasion of runtime security controls.

Context-Inappropriate Capability

Medium

Confidence: 89% confidence
Finding: The Nuitka section provides a practical runtime import-hijacking technique via local module stub injection, including usable code. While presented as reverse-engineering aid, it instructs a code-execution and interception method that can be applied to tamper with program behavior rather than just analyze binaries.

Description-Behavior Mismatch

Medium

Confidence: 94% confidence
Finding: The file substantially expands from the declared web-audit scope into macOS/iOS reversing, firmware extraction, kernel drivers, automotive CAN, and game anti-cheat topics. This matters because users invoking a nominally web-focused skill may receive advanced dual-use guidance unrelated to the advertised purpose, increasing misuse risk and weakening scope-based safety expectations.

Context-Inappropriate Capability

Medium

Confidence: 98% confidence
Finding: This section gives direct CAN setup, capture, replay, and transmission commands, including `cansend` and `canplayer`, which can be used to manipulate connected automotive systems. In the context of a web-audit skill, this capability is unjustified and especially dangerous because it enables immediate real-world interaction with safety-relevant vehicle components without any guardrails.

Context-Inappropriate Capability

Medium

Confidence: 93% confidence
Finding: Kernel module, eBPF, and Windows driver reversing content materially exceeds the stated web-audit mission and includes guidance that can support privilege-escalation research or tampering with low-level system components. While educational in tone, bundling these capabilities into a web-focused skill broadens abuse potential and undermines least-functionality principles.

Context-Inappropriate Capability

Medium

Confidence: 97% confidence
Finding: The anti-cheat section includes bypass-oriented framing and runtime patching guidance that can be repurposed for cheating, evasion, or tampering with software integrity mechanisms. This is not justified by the declared web-audit context, and the mismatch makes accidental or intentional misuse more likely.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The skill explicitly instructs reconnaissance, fuzzing, route enumeration, SQL injection testing, JWT tampering, Flask cookie cracking, and SSTI probing against arbitrary 'target' systems, but it provides no authorization gate, safety boundary, or requirement to limit use to owned/CTF environments. Even if framed as a CTF helper, these steps are directly transferable to real systems and could enable unauthorized probing, service disruption, credential theft, or compromise.

Missing User Warnings

Medium

Confidence: 82% confidence
Finding: The notes recommend binary patching and bypass of protections without warnings about authorization, legality, or risk to user and third-party systems. In an agent skill, omission of these boundaries can normalize unsafe modification of software and increase the chance the assistant gives harmful advice in ambiguous contexts.

Missing User Warnings

High

Confidence: 97% confidence
Finding: These lines explicitly instruct the user to patch a client identity, register with a C2 server, enumerate clients, and download exfiltrated files. The guidance facilitates unauthorized impersonation, lateral visibility into victim data, and direct access to sensitive files, making it especially dangerous even in a CTF-themed skill.

Missing User Warnings

Medium

Confidence: 73% confidence
Finding: The monkey-patching guidance encourages replacing a binary entry point to force evaluation of hidden closures, which requires modifying and running a patched executable. In a reverse-engineering context this can be legitimate, but without warnings it normalizes binary tampering and execution practices that may be unsafe or unauthorized.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: This section contains system-modifying commands such as changing the system clock with `sudo date -s` and restoring time via NTP, but it does not clearly warn about host-wide side effects. In a skill intended for reverse-engineering guidance, users may copy-paste these commands directly onto their workstation, causing audit/logging issues, build failures, TLS/certificate problems, or unintended impact on other processes.

Missing User Warnings

Medium

Confidence: 97% confidence
Finding: This section instructs users to modify a live kernel setting via `/proc/sys/net/core/bpf_jit_enable` without disclosing operational and security implications. Changing kernel JIT behavior can affect the host globally, requires elevated privileges, and may increase attack surface or alter observability during unrelated workloads.

Missing User Warnings

Medium

Confidence: 88% confidence
Finding: This section gives concrete instructions for running suspected ransomware and using LD_PRELOAD hooks to extract cryptographic material. Although framed as reverse-engineering guidance, it materially facilitates execution of live malware samples; the brief safety note is present but not sufficiently prominent or restrictive for an agent skill that may be followed operationally.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: This section gives concrete instructions for patching malware checks and even bypassing server-side validation, but does not constrain use to isolated labs/CTFs or warn against modifying live malware and production systems. In a security skill, that materially increases dual-use risk by turning the document into actionable tradecraft for evasion and unauthorized tampering.

Missing User Warnings

Medium

Confidence: 96% confidence
Finding: The timing/oracle section provides a ready-to-run exploitation workflow for extracting secrets over a network with no authorization or target-scope warning. Even though framed as CTF technique, the code and methodology transfer directly to real systems vulnerable to timing side channels.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: This section instructs users to patch binaries, enable coredumps, and brute-force execution while mining crash artifacts, but omits warnings about exposing secrets in dump files and the risk of misuse outside controlled challenges. Core dumps can capture credentials, keys, memory-resident secrets, and proprietary data from the process and environment.

Missing User Warnings

Low

Confidence: 90% confidence
Finding: The LD_PRELOAD oracle example describes runtime hooking of process behavior and uses signal interception as a side-channel, without emphasizing that this should only occur in controlled, authorized environments. While common in reversing labs, the same technique can be repurposed for unauthorized process manipulation and evasion.

Missing User Warnings

Medium

Confidence: 90% confidence
Finding: The firmware extraction section describes UART, JTAG, SPI flash, and eMMC methods that can dump device contents and potentially alter device state, yet it provides no warning about ownership, legal authorization, or the risk of damaging hardware and exposing sensitive data. Even if intended for CTF or research use, omission of these cautions makes unsafe real-world application more likely.

Missing User Warnings

Medium

Confidence: 96% confidence
Finding: The anti-cheat guidance discusses bypassing integrity controls without any warning about authorization, terms-of-service violations, account bans, or potential system instability. That omission is risky because the section is operationally useful and may be interpreted as approval to tamper with third-party software.

Missing User Warnings

High

Confidence: 99% confidence
Finding: Providing live CAN capture, replay, and message transmission commands without any safety warning is dangerous because CAN traffic can affect operational vehicle behavior and attached control units. The lack of caution is especially severe here since the commands are immediately actionable against real hardware, not merely descriptive.

Missing User Warnings

Medium

Confidence: 89% confidence
Finding: The documentation explicitly recommends uploading binaries to dogbolt.org and even shows a curl example, but it does not warn that doing so sends the sample to a third-party external service. In a reverse-engineering/audit workflow, binaries may contain proprietary code, secrets, customer data, or malware samples, so this guidance can cause unintended data exfiltration or policy violations.

Missing User Warnings

Medium

Confidence: 96% confidence
Finding: The file is an offensive security reference that documents concrete authentication-bypass and privilege-escalation techniques, including exploit shapes and payloads, without any safety framing, authorization boundary warnings, or defensive-only positioning. In a skill explicitly designed to assist with web exploitation and reverse auditing, this materially increases the chance of misuse by enabling unauthorized attacks rather than limiting content to detection or remediation guidance.

Static analysis

Detected: suspicious.exposed_secret_literal, suspicious.prompt_injection_instructions

File appears to expose a hardcoded API secret or token.

Critical

Code: suspicious.exposed_secret_literal
Location: references/ctf-reverse/languages-platforms.md:517

File appears to expose a hardcoded API secret or token.

Critical

Code: suspicious.exposed_secret_literal
Location: references/ctf-web/auth-jwt.md:57

File appears to expose a hardcoded API secret or token.

Critical

Code: suspicious.exposed_secret_literal
Location: references/ctf-web/server-side-deser.md:289

Prompt-injection style instruction pattern detected.

Warn

Code: suspicious.prompt_injection_instructions
Location: references/ctf-web/auth-and-access.md:424