The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze data from untrusted external sources, which presents an attack surface for indirect prompt injection.
Ingestion points: Data enters the agent's context through responses from curl requests to GraphQL endpoints, scanning of JavaScript bundles via grep, and reports generated by external tools like clairvoyance or graphql-cop.
Boundary markers: The skill includes a "Mandatory confirmation gate" requiring the agent to obtain target details and authorization before performing active tests. However, it does not specify technical delimiters (like XML tags or specific markers) to separate untrusted external content from its own instructions.
Capability inventory: The skill utilizes curl for network requests, grep for pattern matching, and various CLI security tools (inql, clairvoyance, graphql-cop) for scanning and exploitation.
Sanitization: There are no instructions for the agent to sanitize, escape, or filter content retrieved from external sources before processing it or incorporating it into the conversation.
[COMMAND_EXECUTION]: The skill provides numerous examples of curl, grep, and python commands to be executed in the terminal for reconnaissance and exploitation of GraphQL services. These are intended for security testing purposes and are preceded by a mandatory authorization check.

hunt-graphql