The Agent Skills Directory

[Data Exposure & Exfiltration] (MEDIUM): The script scripts/ts-api.sh accesses a sensitive credential file at ~/.clawdbot/credentials/tailscale/config.json to retrieve the apiKey. This key is then transmitted via curl to api.tailscale.com. While this is the intended function of a management skill, the exposure of high-privilege keys to an AI agent poses a risk of accidental or malicious disclosure.
[Command Execution] (MEDIUM): The skill enables high-risk network operations via the tailscale CLI, including tailscale funnel (exposing local ports to the public internet) and tailscale up --ssh (enabling remote shell access). These capabilities could be leveraged to bypass local firewall protections or establish unauthorized persistence.
[Indirect Prompt Injection] (LOW): The skill ingests untrusted data from the Tailscale network environment which could potentially contain malicious instructions.
Ingestion points: scripts/ts-api.sh retrieves device lists, hostnames, and ACL policies from the Tailscale API.
Boundary markers: None. Data is parsed via jq and returned as raw or structured text to the agent context.
Capability inventory: The skill can create reusable auth keys, delete devices, and expose services to the internet.
Sanitization: No sanitization or validation is performed on hostnames or device tags retrieved from the API, allowing for potential injection if an attacker can name a device in the tailnet.

tailscale