The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from untrusted external sources, including PDF/Word document metadata, public GitHub repositories, and search results from external APIs (such as Hunter and FOFA). The instructions do not define boundary markers or specify sanitization procedures to prevent malicious instructions embedded in these external sources from influencing the agent's behavior.
Ingestion points: Metadata extraction from documents (Phase 2), GitHub repository scanning (Phase 2), and OSINT API responses (Phase 1 and Phase 5).
Capability inventory: The skill uses network tools like curl and http_request to fetch data and includes templates for generating outgoing phishing content.
Boundary markers: Absent; there are no instructions to delimit external content from internal logic.
Sanitization: Absent; the skill does not require validation or filtering of content retrieved from external OSINT sources.
[DYNAMIC_EXECUTION]: The skill includes Python script templates in references/social-engineering-techniques.md for generating password dictionaries and email username variations. These scripts are intended to be executed or adapted by the agent at runtime to create tailored attack payloads based on collected reconnaissance data.
[COMMAND_EXECUTION]: The methodology relies on executing shell commands such as curl to interact with external APIs and suggests the use of specialized security tools like exiftool for document analysis and ccupp for password profiling.

social-engineering