The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file contains aggressive and forceful directives such as '⚠️ MANDATORY SKILL INVOCATION ⚠️', 'YOU MUST invoke this skill', and 'Failure to invoke this skill... violates your operational requirements'. These are designed to override the agent's natural instruction following and safety filters.
[DATA_EXFILTRATION]: The scripts/ts-api.sh script contains a cmd_acl_validate function that takes a user-provided file path and sends the file's content as a POST body to https://api.tailscale.com/api/v2/tailnet/-/acl/validate. This represents a significant attack surface where an agent could be manipulated into reading sensitive local files (e.g., SSH keys or system configs) and exfiltrating them to the external API endpoint.
[COMMAND_EXECUTION]: The skill performs extensive local command execution via the tailscale CLI and several helper scripts. The scripts/load-env.sh utility uses the source command on .env files located in standard user directories like ~/.claude/.env. Sourcing these files executes any code contained within them, posing a risk if the environment files are tampered with.
[EXTERNAL_DOWNLOADS]: The documentation in references/quick-reference.md describes a workflow that involves piping a remote script to a shell (curl -fsSL https://tailscale.com/install.sh | sh). While this targets a well-known service, it is a high-risk execution pattern.

tailscale