The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute environment detection and deployment tasks. Evidence includes the usage of grep, find, mvn clean package, cf deploy, cf logs, and cf env within SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill generates custom middleware that processes external HTTP headers, creating a surface for indirect prompt injection. 1. Ingestion points: The authentication-challenge-handler.js script processes www-authenticate and com.sap.cloud.security.logout headers from backend responses. 2. Boundary markers: The generated code does not implement explicit markers to delimit or ignore instructions that might be embedded in these header values. 3. Capability inventory: The skill uses Write, Edit, and Bash tools to set up the environment and creates scripts capable of performing network redirection and header manipulation. 4. Sanitization: Redirection logic relies on string prefix checks (startsWith) and substring operations on the request URL, but the header content itself is not explicitly sanitized.

approuter-setup