The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to clone a repository from an unverified GitHub account (github.com/EastSword/skill-dfyx_code_security_review.git) and install dependencies via pip install -r requirements.txt from that same source.
[REMOTE_CODE_EXECUTION]: The documentation describes a 'Docker-Based Verification' feature designed to automatically execute exploit payloads (PoCs) to confirm vulnerabilities, which constitutes remote code execution when triggered by audited content.
[COMMAND_EXECUTION]: The skill relies on several helper scripts (code_scan.py, vulnerability_validator.py, pattern_scanner.py) that perform direct shell command execution and environment variable manipulation.
[PRIVILEGE_ESCALATION]: The configuration explicitly mentions using DOCKER_HOST=unix:///var/run/docker.sock, which grants the skill high-privileged access to the host system's Docker daemon, potentially allowing for container escapes or host compromise.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted project source code, creating a surface for indirect injection attacks.
Ingestion points: Project source files (Java, Python, Go, etc.) provided by the user for auditing.
Boundary markers: None identified; instructions do not specify delimiters to separate code content from analysis instructions.
Capability inventory: Shell command execution, Docker socket interaction, network requests to external APIs (NVD, GitHub), and local file system writes.
Sanitization: While the skill methodology mentions 'Taint Tracking', there are no explicit instructions for the agent to sanitize or ignore instructions embedded within the source code files it analyzes.

dfyx-code-security-auditor