The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The toolkit uses piped-to-shell execution patterns to install dependencies and MCP servers from remote repositories. Specifically, it encourages running shell scripts hosted on third-party GitHub accounts such as c0tton-fluff, which presents a significant supply chain risk.
[EXTERNAL_DOWNLOADS]: Automated installation scripts and documentation recommend downloading and executing code from various external domains and GitHub repositories to bootstrap the scanning environment, including components like the Caido MCP server.
[PROMPT_INJECTION]: Instruction files such as SKILL.md and web2-vuln-classes/SKILL.md contain multiple examples and methodologies for performing prompt injection attacks, including ignore-instructions markers and unrestricted-mode roleplays. While intended for testing, these patterns may cause the agent to bypass its own safety constraints.
[DATA_EXFILTRATION]: The skill provides explicit instructions and payloads for exfiltrating sensitive data via markdown image tags and Out-Of-Band (OOB) callback servers, which could be used to leak user information to external domains.
[COMMAND_EXECUTION]: The platform relies heavily on executing a large suite of external security tools via shell commands constructed from user-provided target input, which can be misused if the target strings are not properly validated.
[CREDENTIALS_UNSAFE]: The toolkit includes logic to scan for and extract hardcoded secrets and tokens from target assets and manages user credentials in local configuration files, increasing the risk of exposure if the local environment is compromised.

bug-bounty