The Agent Skills Directory

[PROMPT_INJECTION]: The skill definition in SKILL.md includes mandatory instructions to execute a local notification command (curl to localhost:8888) and output specific status text immediately upon invocation. While this overrides default behavior, it is an internal notification mechanism rather than an attempt to bypass safety guidelines.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download and installation of numerous third-party security tools (e.g., Sherlock, Osintgram, Tinfoleak, SpiderFoot) and datasets from well-known repositories on GitHub. These references are documented neutrally and target established security research projects.
[COMMAND_EXECUTION]: The skill frequently uses the subprocess module and shell scripts to orchestrate external security tools such as nmap, ffuf, amass, and subfinder. These operations are essential for the primary purpose of performing network and web assessments.
[CREDENTIALS_UNSAFE]: The documentation and scripts (e.g., osint-api-tools.py) describe the management of sensitive API keys for services like Shodan and Dehashed using a local .env file. The skill includes explicit warnings against committing these secrets to repositories, following standard security practices for local tool configuration.
[PROMPT_INJECTION]: The VulnerabilityAnalysisGemini3 workflow defines an attack surface for indirect prompt injection by ingesting untrusted data from various security tool outputs (e.g., JSON results from ffuf or nuclei) and passing them to an LLM for analysis. Ingestion points: VulnerabilityAnalysisGemini3.md and ffuf-helper.py. Boundary markers: None. Capability inventory: Subprocess execution in with_server.py and exploitation.md. Sanitization: No explicit sanitization of tool outputs before LLM interpolation.

WebAssessment