The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill contains a pattern that pipes data from a remote URL directly into a Python interpreter (curl https://target.com/.well-known/openid-configuration | python3 -m json.tool). While used here for JSON formatting, this represents a high-risk remote code execution vector common in malicious scripts.
[COMMAND_EXECUTION]: Instructions include the use of adb shell and apktool to interact with and decompile Android applications, enabling command execution on connected mobile devices and local file system manipulation.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to external, user-defined targets to fetch OIDC configuration files, social login endpoints, and registration metadata.
[OBFUSCATION]: Payloads include URL encoding (%2F, %40, %23) and double encoding (%252F) to evade redirect_uri filters. It also references homograph techniques (šemrush.com) to deceive validation logic.
[PRIVILEGE_ESCALATION]: Use of adb shell provides high-level access to Android device internals, which could be used to bypass application sandbox protections during OAuth hunting.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection. Ingestion points: It processes external data from discovery endpoints and JS bundles. Capability inventory: It has network (curl), system-level (adb), and file system (apktool) capabilities. Boundary markers: No delimiters or warnings are used for processed data. Sanitization: No input filtering or escaping is implemented for content retrieved from targets.

hunt-oauth