claude-code-cybersecurity-skill
Fail
Audited by Gen Agent Trust Hub on Jun 22, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill documentation instructs users to clone a remote repository from an unverified source (
https://github.com/Masriyan/Claude-Code-CyberSecurity-Skill.git) to obtain the full collection of security skills. - [REMOTE_CODE_EXECUTION]: By guiding the user to install 15 unverified
SKILL.mdfiles and associated Python automation scripts from the external repository into the agent's global skill directory (~/.claude/skills/), the skill establishes a mechanism for unvetted remote code and instructions to be executed by the agent. - [COMMAND_EXECUTION]: The skill methodologies and examples frequently involve the agent executing potentially dangerous shell commands (e.g.,
nmap,subfinder) and custom scripts (scripts/harden_linux.py,scripts/vt_lookup.py) via the bash tool. - [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8) because its primary purpose is to ingest and analyze untrusted external data (phishing emails, malware reports) while possessing high-privilege capabilities like shell access, without defining explicit boundary markers or sanitization procedures.
- Ingestion points: Phishing emails, malware report PDFs, and suspicious PE files processed via the agent's file reading tools.
- Boundary markers: None present; the methodology instructions do not specify the use of delimiters or warnings to ignore embedded instructions in the processed data.
- Capability inventory: Full bash shell access, Python script execution, and file system write capabilities across all scripts mentioned (e.g.,
scripts/harden_linux.pyinSKILL.md). - Sanitization: No evidence of input validation, escaping, or filtering of external content before processing.
Recommendations
- AI detected serious security threats
Audit Metadata