The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill involves the ingestion and processing of untrusted external data, such as DNS records and HTTP error messages, to identify vulnerabilities. This represents a potential surface for indirect prompt injection if the external data contains instructions designed to mislead the agent.
Ingestion points: The agent is instructed to parse output from DNS resolution tools like dig and HTTP clients like curl (referenced in SKILL.md Sections 2, 4, and 5) as well as various subdomain enumeration tools.
Boundary markers: The instructions do not define clear delimiters or "ignore embedded instructions" warnings when processing the output from external network probes.
Capability inventory: The skill context includes access to infrastructure management tools (AWS CLI, Heroku CLI, Git) and network diagnostics, which are powerful capabilities if redirected by malicious input.
Sanitization: There are no explicit instructions to sanitize or validate the content of HTTP fingerprints or DNS records before the agent uses them to make decisions about resource claiming.

subdomain-takeover