The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides Python code snippets in references/web-and-dns.md that programmatically access and read browser history files (e.g., sqlite3.connect("History")) to search for Telegram bot interactions. This access to sensitive personal data is presented as a digital forensic technique for CTF challenges.
[EXTERNAL_DOWNLOADS]: The instructions reference numerous external APIs and services for information gathering, including ip-api.com, ipinfo.io, web.archive.org, securitytrails.com, and whoisxmlapi.com. Most of these are well-known services utilized for security research and reconnaissance.
[COMMAND_EXECUTION]: The documentation includes various command-line instructions using tools like dig, exiftool, curl, and grep to perform DNS reconnaissance, metadata extraction, and local file searches. These are intended for user-guided execution during investigative tasks.
[INDIRECT_PROMPT_INJECTION]: The skill lacks explicit sanitization or boundary markers when processing untrusted external data such as image metadata, document contents, or DNS records. This presents an attack surface where malicious data embedded in external sources could influence the agent's behavior during analysis.
Ingestion points: references/web-and-dns.md (browser history files), references/geolocation-and-media.md (image EXIF and document metadata).
Boundary markers: Absent from processing instructions.
Capability inventory: curl (network operations), exiftool (file metadata extraction), dig (network DNS queries).
Sanitization: No evidence of sanitization or validation for tool outputs before their use in subsequent analysis steps.

ctf-osint