The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Automated scanners have identified the documentation URL https://jeffallan.github.io/claude-skills/skills/security/security-reviewer/ as a blacklisted malicious domain. This URL is integrated into the skill's metadata and reference documentation.
[METADATA_POISONING]: The SKILL.md file has been flagged by file reputation scanners as malicious (FileRepMalware).
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data (such as code and logs) while possessing access to high-privilege tools like Bash. It lacks input sanitization or clear boundary markers, meaning malicious instructions within analyzed files could potentially control the agent's actions.
Ingestion points: Source code, infrastructure manifests, and log files read via Read, Grep, and Glob tools.
Capability inventory: System command execution via the Bash tool for running scanners and exploiting vulnerabilities.
Boundary markers: None identified in the instruction set.
Sanitization: No validation or escaping of external content is specified.
[COMMAND_EXECUTION]: The skill provides numerous commands for active reconnaissance and exploitation validation, including nmap, sqlmap, and various cloud platform CLI tools, which can be leveraged for malicious purposes if the agent is compromised.
[DATA_EXFILTRATION]: The skill provides documentation and patterns (e.g., regex for AWS keys, private keys, and JWTs) specifically for locating and extracting sensitive credentials. The presence of a blacklisted domain in the documentation links poses a risk of these credentials being transmitted to untrusted infrastructure.

security-reviewer