The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a wide range of external security tools including nmap, amass, subfinder, nuclei, feroxbuster, and ffuf via shell commands. These tools are invoked with user-supplied arguments such as $DOMAIN, $TARGET, and $EMAIL.
[DATA_EXFILTRATION]: Target-specific information (domains, IPs, email addresses) is transmitted to multiple external OSINT services and APIs, including crt.sh, nvd.nist.gov, api.github.com, haveibeenpwned.com, and Shodan. While intended for reconnaissance, this pattern involves sending data to third-party endpoints.
[EXTERNAL_DOWNLOADS]: The skill fetches content and data from numerous external sources, such as the Wayback Machine, GitHub's API, and the National Vulnerability Database (NVD).
[CREDENTIALS_UNSAFE]: The skill instructs the agent to use an environment variable $HIBP_API_KEY directly within a curl command header. This can lead to the API key being exposed in shell history, process listings, or logs on the host system.
[PROMPT_INJECTION]: There is a surface for indirect prompt injection (Category 8) as the skill ingests and processes untrusted data from external sources (e.g., GitHub repository descriptions, Wayback Machine URLs) which are then parsed and included in reports.
Ingestion points: SKILL.md (Phase 1, 2, and 3 via curl, waybackurls, etc.)
Boundary markers: Absent; external data is processed and formatted without explicit delimiters or instructions to ignore embedded commands.
Capability inventory: Extensive command execution capability across all phases in SKILL.md.
Sanitization: Uses jq for basic JSON structure parsing, but the resulting text is not sanitized against secondary injection attacks.

recon-osint