The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process a wide variety of untrusted data from external sources, including breach logs, DNS records, WHOIS data, and web scrapers. This creates a significant surface for indirect prompt injection attacks where malicious instructions embedded in OSINT data could influence the agent's behavior.
Ingestion points: Data enters the context via API responses (e.g., HudsonRock in references/breach-and-credentials.md), web scraping (e.g., references/people-osint.md), and DNS query results.
Boundary markers: The skill lacks explicit instructions for the agent to isolate or treat ingested data as untrusted content, though it includes scoring rubrics in SKILL.md to evaluate asset confidence.
Capability inventory: The skill leverages tool execution capabilities including curl, dig, whois, and python scripts to interact with the environment and external services.
Sanitization: No specific sanitization or filtering logic is prescribed for the raw data ingested from external OSINT sources.
[EXTERNAL_DOWNLOADS]: The references/tooling-install.md file contains instructions for downloading and installing a large number of third-party security tools from GitHub and various package registries (npm, pip, cargo, etc.). While many of these tools are well-known in the security community (e.g., Subfinder, Amass, Nuclei), they originate from diverse third-party maintainers.
[COMMAND_EXECUTION]: The skill heavily relies on the execution of shell commands and one-liners for reconnaissance tasks. Numerous examples in references/probes-and-wordlists.md and references/identity-fabric.md demonstrate the use of curl, dig, and grep to perform network probes and data extraction.
[PRIVILEGE_ESCALATION]: In references/tooling-install.md, the installation instructions for the masscan tool require the use of sudo, which is a privileged operation.

offensive-osint