The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands using the curl tool. The logic interpolates a user-provided variable ($DOMAIN) directly into the command string: curl ... "https://$DOMAIN/". This creates a vulnerability where malicious input containing shell metacharacters (e.g., ;, &, |, backticks) could execute arbitrary commands on the system.
[DATA_EXFILTRATION]: The skill performs outbound network requests to user-specified domains to retrieve HTTP headers. This is a core part of its functionality but constitutes communication with untrusted external servers.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted content from the web.
Ingestion points: Data enters the agent's context via the output of curl commands stored in /tmp/headers.txt (SKILL.md).
Boundary markers: No boundary markers or clear delimiters are used to isolate the untrusted header data from the agent's instructions.
Capability inventory: The skill has access to curl and grep for file and network operations (SKILL.md).
Sanitization: No sanitization or validation is performed on the ingested header content before it is processed by the agent.

security-headers-check