The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill's operational instructions in SKILL.md explicitly direct the use of elevated privileges via sudo to execute the phishing framework (e.g., sudo ./bin/evilginx -p ./phishlets).
[COMMAND_EXECUTION]: The skill workflow involves shell command execution for installing and operating tools. Specifically, SKILL.md includes commands for building the framework with make and executing it. Furthermore, scripts/agent.py uses subprocess.run to execute evilginx --version to verify the environment.
[EXTERNAL_DOWNLOADS]: The skill workflow requires downloading code from external GitHub repositories that are not on the trusted list, including https://github.com/kgretzky/evilginx2.git and https://github.com/fin3ss3g0d/evilgophish.
[OBFUSCATION]: The instructions in SKILL.md (Phase 1) recommend the registration of lookalike domains using homoglyphs or typosquatting to create deceptive landing pages for targets.
[INDIRECT_PROMPT_INJECTION]: The skill's analysis scripts present a surface for indirect prompt injection through the ingestion of external data.
Ingestion points: scripts/agent.py reads and parses external YAML phishlet files and session logs; scripts/process.py processes raw session output.
Boundary markers: None identified in the processing scripts to separate untrusted data from instructions.
Capability inventory: scripts/agent.py contains capability for subprocess command execution.
Sanitization: While scripts/agent.py uses yaml.safe_load() to prevent unsafe deserialization of phishlets, it does not sanitize the content of the logs or phishlets against potential prompt injection attacks within the agent's context.

performing-initial-access-with-evilginx3