Security audit

CTF Misc & Forensics

Security checks for vulnerabilities and agentic risk

Overview

This CTF helper is mostly a visible reference corpus, but it includes powerful attack, credential-access, and account-action workflows that go well beyond its forensics-focused description.

Install only if you want a broad CTF playbook and will keep it limited to authorized competitions or isolated labs. Do not let an agent use these references against real systems, local hosts, cloud buckets, credentials, or CTF accounts without explicit user confirmation and clear authorization.

SkillSpector

Vulnerability Patterns

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
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (100)

Context-Inappropriate Capability

High

Confidence: 97% confidence
Finding: This section includes active attack guidance for Timeroasting, including extracting crackable hashes and running hashcat, which exceeds passive forensic analysis and enables credential compromise techniques. In a skill with allowed Bash/Web access, this materially increases the risk of misuse for unauthorized offensive operations.

Context-Inappropriate Capability

Medium

Confidence: 94% confidence
Finding: The cloud storage section provides active remote enumeration and object retrieval commands for AWS S3, GCP, and Azure, which extends beyond passive local disk/memory forensics into live access against external services. In a skill marketed for misc/pcap/file/memory/disk forensics, this broadens operational capability to potentially unauthorized data access and increases abuse potential.

Context-Inappropriate Capability

Medium

Confidence: 97% confidence
Finding: The MD5 PDF collision section teaches construction of colliding files, which is a cryptographic exploitation technique unrelated to disk or memory forensics. Even in CTF context, it introduces a dual-use capability that can be repurposed for integrity bypasses, deception, or document substitution outside the declared skill scope.

Context-Inappropriate Capability

Medium

Confidence: 97% confidence
Finding: This section goes beyond defensive forensics and documents how to recover DNSSEC private keys from git history and then use them to forge DNSSEC-signed responses. That materially enables offensive abuse against real zones if reused outside a CTF, and it is misaligned with the skill's stated forensics/misc recovery purpose, which increases risk because an agent could surface it in inappropriate contexts.

Context-Inappropriate Capability

Medium

Confidence: 90% confidence
Finding: This section introduces live external-service querying and transaction tracing against public blockchain infrastructure, which goes beyond passive/offline forensics. In a general-purpose agent skill, that broadens capability from artifact analysis into active reconnaissance and network egress, which can be misused and may violate containment expectations.

Context-Inappropriate Capability

Medium

Confidence: 95% confidence
Finding: The KeePass section provides actionable credential extraction, transfer, and cracking workflows, including wordlist generation and hashcat usage. That materially exceeds simple forensic identification/restoration and enables credential recovery that could be abused for unauthorized access.

Context-Inappropriate Capability

Medium

Confidence: 92% confidence
Finding: This section includes executable code that performs active DNS queries against a live target domain rather than restricting itself to offline PCAP analysis. In a forensics skill, that crosses into external interaction and can cause unauthorized probing or be copied into real-world use without approval.

Intent-Code Divergence

High

Confidence: 95% confidence
Finding: The prose says only pixel increments of exactly 1 should encode a 1 bit, but the code maps any pixel change to 1. In forensic or CTF workflows this can silently produce false decodes, mislead analysts, and cause incorrect conclusions or wasted effort when processing attacker-controlled media.

Description-Behavior Mismatch

High

Confidence: 96% confidence
Finding: The skill advertises itself as a miscellaneous fallback, yet it embeds privilege-escalation, container escape, and host-breakout content unrelated to benign puzzle-solving. In an agent context with bash and filesystem access, this materially expands the attack surface by normalizing post-exploitation techniques that could be applied outside intended CTF sandboxes.

Context-Inappropriate Capability

High

Confidence: 99% confidence
Finding: This section contains actionable privilege-escalation and breakout procedures, including sudo abuse, Docker-to-host chroot, SUID abuse, and service misconfiguration exploitation. In a tool-using agent, these instructions could directly enable takeover of the local machine, containers, or adjacent systems if the agent is ever pointed at a real environment rather than an isolated challenge.

Context-Inappropriate Capability

Medium

Confidence: 88% confidence
Finding: The skill provides authenticated API interaction and submission workflows using bearer-style CTF tokens. While legitimate in CTF play, embedding token-driven remote actions in a broad misc skill increases the chance an agent will solicit secrets and perform network actions beyond its core analytical purpose.

Description-Behavior Mismatch

High

Confidence: 97% confidence
Finding: This section explicitly provides a post-compromise privilege-escalation checklist, including SUID and capability discovery, which materially exceeds the declared misc/forensics scope. In a general agent skill, this enables offensive guidance after shell escape and increases the chance the assistant will help users escalate access on real systems.

Description-Behavior Mismatch

High

Confidence: 98% confidence
Finding: This section includes direct exfiltration and reverse-shell techniques using /dev/tcp, including commands to send files to an attacker host and establish bidirectional shell access. That is plainly inconsistent with a forensics assistant and provides immediately actionable offensive tradecraft.

Context-Inappropriate Capability

Critical

Confidence: 99% confidence
Finding: The LD_PRELOAD section teaches users to compile a malicious shared object with a constructor that executes privileged commands, then inject it into an allowed binary to bypass rbash restrictions. This is a concrete code-execution bypass and privilege-abuse technique with real-world applicability, not merely educational background for misc/forensics tasks.

Description-Behavior Mismatch

Medium

Confidence: 91% confidence
Finding: The document substantially expands the skill from forensics/misc analysis into full CTF platform operation, including authentication, challenge enumeration, downloads, submissions, and team/account interactions. That scope drift is risky because it enables live interaction with third-party competition infrastructure and user accounts beyond the manifest’s declared purpose, increasing the chance of unauthorized or policy-violating actions.

Context-Inappropriate Capability

Medium

Confidence: 93% confidence
Finding: These sections document active competition-account actions such as flag submission and paid hint unlocking, which can spend points, alter user/team state, and interact with external systems. For a forensics-analysis skill, this is unnecessarily dangerous because it encourages execution of authenticated actions rather than analysis-only assistance.

Context-Inappropriate Capability

High

Confidence: 96% confidence
Finding: This section contains operational offensive guidance for DNS rebinding, including attack flow, sample server code, and references to tooling that can be used against real systems. In a skill घोषित as misc/forensics assistance, this materially expands the agent into active exploitation, increasing the chance it facilitates unauthorized access rather than analysis.

Context-Inappropriate Capability

Medium

Confidence: 91% confidence
Finding: The round-robin enumeration section instructs repeated DNS collection, port scanning, and direct virtual-host probing across discovered IPs. Those are active recon techniques outside the stated forensics scope and could be misused to probe third-party infrastructure.

Context-Inappropriate Capability

Critical

Confidence: 99% confidence
Finding: This is explicit exploitation content for TCP Fast Open SYN-payload command injection, including runnable code that sends payloads in SYN packets and an example command reading a secret file. It enables stealthy command execution against services that may evade normal scans, making it highly dangerous and far outside misc/forensics analysis.

Context-Inappropriate Capability

Medium

Confidence: 95% confidence
Finding: This section provides concrete exploit instructions and code for abusing a remote Ruby service to trigger CVE-2018-8778 and dump memory from a target. That goes beyond passive forensics or encoding assistance and materially enables offensive exploitation, making misuse straightforward even if presented in a CTF context.

Description-Behavior Mismatch

High

Confidence: 96% confidence
Finding: The file contains step-by-step instructions and code for bypassing game logic, manipulating session state, abusing APIs, and patching runtime memory, which materially exceeds the stated scope of a misc/forensics assistant. Even if framed as CTF content, these are transferable offensive techniques that could be applied to real systems with similar design flaws.

Context-Inappropriate Capability

High

Confidence: 94% confidence
Finding: This section teaches browser-console and WebSocket-based client-state manipulation, including directly changing coordinates and invoking validation functions. That is actionable exploitation guidance rather than forensic analysis, and it can be repurposed for cheating, bypassing controls, or probing insecure client/server trust boundaries in live applications.

Context-Inappropriate Capability

High

Confidence: 95% confidence
Finding: The content provides a concrete API abuse pattern: start a session, wait out server-side timing checks, then directly call win/collect endpoints. This is a classic validation-bypass technique that enables unauthorized state transitions whenever a service validates elapsed time but not actual behavior.

Context-Inappropriate Capability

High

Confidence: 97% confidence
Finding: The WASM section gives direct runtime memory patching instructions with example offsets and code to force favorable state and trigger a win. This is operational tampering guidance that can be adapted to manipulate application behavior in any client-executed WASM target, not merely analyze artifacts.

Description-Behavior Mismatch

High

Confidence: 99% confidence
Finding: This section contains explicit offensive exploitation content, including code-jail escape, ROP/shellcode construction, BuildKit secret exfiltration, and Docker/container breakout techniques. For a skill scoped as misc/forensics assistance, this materially increases the chance the agent will provide actionable intrusion guidance rather than defensive analysis.

Static analysis

Detected: suspicious.exposed_secret_literal

File appears to expose a hardcoded API secret or token.

Critical

Code: suspicious.exposed_secret_literal
Location: references/ctf-forensics/linux-forensics.md:108