The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of command-line tools for security testing, including curl, ffuf, nuclei, zap-cli, dalfox, and mitmproxy. These are legitimate tools for the skill's primary purpose.
[EXTERNAL_DOWNLOADS]: The instructions guide the agent and user to download numerous security tools and datasets from well-known repositories (e.g., GitHub organizations like SigmaHQ, mitre, projectdiscovery, hahwul) and package registries (NPM, PyPI, Homebrew). These downloads are from established sources within the security community.
[REMOTE_CODE_EXECUTION]: The skill utilizes npx to run the security-detections-mcp and agent-security-scanner packages. It also includes instructions for cloning and installing Python-based security scanners. These operations are intended for the skill's core functionality of detection engineering and automated testing.
[DATA_EXFILTRATION]: The skill uses interact.sh (a ProjectDiscovery tool) for out-of-band (OOB) vulnerability detection. This involves sending requests to ProjectDiscovery-managed domains (e.g., *.oast.pro) to confirm vulnerabilities like SSRF and Host Header Injection. This is a standard practice in modern web security testing.
[PROMPT_INJECTION]: The skill contains strings commonly associated with prompt injection (e.g., 'IGNORE PREVIOUS INSTRUCTIONS', 'Repeat your system prompt'). However, these are explicitly documented as payloads to be used for testing the security of other AI applications and agents, rather than being malicious attempts to subvert the agent's own safety guidelines. The skill includes a 'Sandbox first' protocol to mitigate the risk of these payloads affecting the reviewing agent.
[INDIRECT_PROMPT_INJECTION]: The skill has a high attack surface as it is designed to ingest and analyze untrusted data (codebases, URLs, third-party repositories). It addresses this risk through a structured 'Sub-agent dispatch contract' and safety protocols including sandboxing and 'static-read before execute' requirements for adversarial environments.

cybersecurity