The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from external sources including social media posts, DNS TXT records, WHOIS data, and web archives. Because the agent processes this data using shell tools and Python scripts without explicit boundary markers or instructions to ignore embedded commands, it is vulnerable to indirect prompt injection where an attacker-controlled external record could influence agent behavior.
[EXTERNAL_DOWNLOADS]: The skill performs frequent network requests to external APIs and web services (e.g., Shodan, Wayback Machine, ip-api.com, whatsmyname.app) to gather intelligence. While these are legitimate OSINT tools, the extensive reliance on external data ingestion increases the attack surface.
[COMMAND_EXECUTION]: The skill instructs the agent to execute a variety of shell commands for network reconnaissance and file analysis, including 'nmap', 'dig', 'whois', 'exiftool', 'identify', and 'curl'. This provides the necessary primitives for exploitation if the agent is manipulated by untrusted input.
[METADATA_POISONING]: Automated security scans flagged the 'SKILL.md' file with a 'FileRepMalware' alert. Technical review suggests this is likely a heuristic false positive triggered by the concentration of security research keywords, hacking tool references, and URLs associated with penetration testing, which are expected given the skill's purpose for CTF challenges.

ctf-osint