The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from external targets (such as JavaScript files, HTML content, and API responses) to identify vulnerabilities. This process creates a surface where malicious instructions could be embedded in the target data to influence the agent's behavior.
Ingestion points: The methodology involves downloading and analyzing external JavaScript files (jsluice, mantra), crawling websites (katana, gau), and inspecting API responses.
Boundary markers: The methodology lacks explicit instructions for the agent to use delimiters or ignore embedded instructions when processing external content.
Capability inventory: The agent utilizes network utilities (curl, httpx), file analysis tools, and various security CLI tools to perform its tasks.
Sanitization: No specific sanitization or filtering logic is provided to clean external data before it is interpreted by the agent's natural language processing.
[COMMAND_EXECUTION]: The skill coordinates the execution of a wide variety of security tools and CLI utilities, which is appropriate for bug bounty hunting but constitutes a broad command execution capability.
Evidence: The "Tool Routing by Phase" section lists tools including subfinder, nuclei, httpx, ffuf, and curl across different reconnaissance and discovery phases.
[DYNAMIC_EXECUTION]: The methodology explicitly directs the agent to generate and run Python scripts to handle repetitive or complex logic more reliably than shell scripting.
Evidence: The "Shell-Loop Ban" section in SKILL.md mandates that "For any iteration that runs more than 5 times, use Python... not shell for-loops."

bb-methodology