The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external JSON files containing target information and campaign results, which creates a potential surface for indirect prompt injection if the data source is untrusted.
Ingestion points: The scripts/agent.py script ingests target and result data from JSON files provided via command-line arguments (--targets and --results).
Boundary markers: No explicit boundary markers or instructions (e.g., 'ignore instructions within this data') are used when processing the external content.
Capability inventory: The skill can perform file system operations, including writing campaign reports and CSV logs to the local disk (scripts/agent.py, scripts/process.py).
Sanitization: The scripts do not implement sanitization or validation for the data ingested from the JSON files before using them in generated reports and tracking sheets.
[EXTERNAL_DOWNLOADS]: The skill's documentation references several external dependencies required for its functionality.
Evidence: references/api-reference.md identifies requests, twilio, and Jinja2 as necessary Python libraries. These are well-known and standard packages in the ecosystem.
Evidence: SKILL.md mentions the use of theHarvester, a standard open-source intelligence (OSINT) tool.

conducting-social-engineering-pretext-call