The Agent Skills Directory

[OBFUSCATION]: The documentation includes a Base64-encoded string within a JavaScript template for HTML smuggling. When decoded, this string represents the header of a Windows executable ('MZ'), illustrating how to deliver binary payloads while evading email gateway detection.
Evidence: references/workflows.md contains the pattern var payload = "TVqQAAMAAAAEAAAA...";.
[DYNAMIC_EXECUTION]: The skill provides a workflow for 'HTML Smuggling,' which uses client-side JavaScript to dynamically construct a binary file from encoded data and trigger a download at runtime.
Evidence: references/workflows.md contains a script using atob(), Uint8Array, and Blob() to reconstruct an executable in the browser.
[INDIRECT_PROMPT_INJECTION]: The skill processes recipient information from external CSV files and interpolates it into email templates using Python string formatting and Jinja2 templates without explicit sanitization, creating an attack surface for malicious inputs in the source data.
Evidence: scripts/process.py reads targets.csv and uses .format() on template bodies; references/api-reference.md mentions Jinja2 usage.
Ingestion points: scripts/process.py (line 342).
Capability inventory: requests network operations in agent.py and process.py; dns.resolver in process.py.
Boundary markers: None present.
Sanitization: None detected.
[EXTERNAL_DOWNLOADS]: The workflow documentation instructs the user to download external software from a well-known source to set up the phishing infrastructure.
Evidence: references/workflows.md includes wget https://github.com/gophish/gophish/releases/latest/download/gophish-v0.12.1-linux-64bit.zip.
[COMMAND_EXECUTION]: The skill contains workflows that instruct the agent or operator to execute shell commands for infrastructure setup, including SSL certificate generation and domain administration.
Evidence: references/workflows.md contains openssl, unzip, and opendkim-genkey command-line instructions.

conducting-spearphishing-simulation-campaign