The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references the tailcat tool, a userspace WireGuard tunnel implementation, hosted at the official Tailscale GitHub repository. It specifically pins a verified commit hash (88929418b1a3f3c74904a3136d6a9e87b1b5b9bb) to be used when the tool is installed or updated via the irinse utility, ensuring version integrity for experimental components.
[INDIRECT_PROMPT_INJECTION]: The skill's core functionality involves ingesting local data and exposing it via a network tunnel, which presents a potential surface for prompt injection or unauthorized data exposure. The skill mitigates these risks through the following evidence chain:
Ingestion points: The agent is instructed to read a specific local file or directory as defined by the user task.
Boundary markers: The instructions mandate pinning exact resources, capturing pre-state configuration for rollback verification, and confirming that no content outside the requested boundary is reachable.
Capability inventory: The skill leverages tailscale serve and tailcat for networking, alongside curl for verifying resource availability.
Sanitization: The agent is explicitly commanded to reject sensitive paths, including credential stores, SSH keys, home directories, and version control repositories, and must use the most restrictive private transport available.

fihan