The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill enables the creation of public tunnel URLs, creating a surface for untrusted external data to be ingested by the agent if it processes responses or webhooks arriving via the tunnel.
Ingestion points: External requests to the public tunnel URL from webhooks, public pages, or remote parties as described in SKILL.md.
Boundary markers: The skill mandates explicit user authorization and warns about provider data access in SKILL.md, but does not provide specific delimiters to separate external tunnel traffic from agent instructions.
Capability inventory: Instructions in SKILL.md involve executing CLI tools like cloudflared, ngrok, and tmole to manage network traffic.
Sanitization: There are no explicit instructions for sanitizing or filtering data received via the tunnels before it is processed by the agent.
[REMOTE_CODE_EXECUTION]: Reference documentation in references/tools.md identifies an installation method for tmole that uses curl piped to sudo bash. This is a standard installation pattern for the service and is accompanied by a requirement to review the script and obtain user permission.
[PRIVILEGE_ESCALATION]: The tmole installation process described in references/tools.md may involve the use of sudo to install the utility.
[COMMAND_EXECUTION]: The skill instructs the agent to run various command-line utilities, including cloudflared, ngrok, tmole, and curl, for establishing and testing network connectivity.
[DATA_EXFILTRATION]: The skill correctly identifies that tunneling traffic through third-party providers involves TLS termination by the provider, which could expose sensitive data such as cookies and authorization tokens, as highlighted in the safety section of SKILL.md.

vp-tunnel