The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The 'Receive a webhook callback' recipe instructs the agent to set up a Python listener that logs external HTTP request headers and bodies to /tmp/webhook-hits.log. This log file acts as an ingestion point for untrusted data from the public internet. If the agent is subsequently instructed to process these logs to make decisions or execute further commands, it becomes vulnerable to indirect prompt injection from the external caller. No sanitization or boundary markers are defined for processing this log content.
[DYNAMIC_EXECUTION]: The skill uses a 'heredoc' pattern to generate a temporary Python script (/tmp/webhook-server.py) at runtime and then executes it using nohup python3. While the template is static within the skill, runtime code generation and execution is a dynamic pattern that requires oversight.
[COMMAND_EXECUTION]: The skill makes heavy use of the terminal tool to execute shell commands, manage background processes (nohup, &), and manage process IDs (.pid files). It specifically recommends disabling SSH host key verification (-o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null), which facilitates automated connections but removes protection against machine-in-the-middle (MITM) attacks for the tunnel connection.

pinggy-tunnel