The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documentation provides shell commands for installing, configuring, and running Twingate CLI tools and the Twingate Client.
Evidence: Commands like twingate setup --headless=-, tgcli auth login, and deno compile are detailed in the reference corpus.
[EXTERNAL_DOWNLOADS]: Instructions include downloading packages and scripts from vendor-controlled domains.
Evidence: Downloads are sourced from packages.twingate.com, binaries.twingate.com, and official GitHub repositories under the Twingate-Labs and Twingate-Solutions organizations.
[REMOTE_CODE_EXECUTION]: The documentation includes patterns for downloading and immediately executing setup scripts from the vendor's infrastructure.
Evidence: A setup script for the Twingate Connector is fetched from binaries.twingate.com and piped directly into bash for installation.
[PRIVILEGE_ESCALATION]: The skill documents the use of sudo for system-level tasks such as package management and starting network services.
Evidence: Commands such as sudo apt install twingate, sudo twingate start, and sudo tee /etc/apt/sources.list.d/twingate.list are provided for environment setup.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external data sources, which could potentially contain malicious instructions intended to influence the agent's behavior.
Ingestion points: The skill ingests Twingate Network Events (CSV files) and AWS resource tags (tg_resource, tg_groups).
Boundary markers: The skill includes specific guidelines for validating API responses (checking ok fields) and emphasizes least-privileged token management, but lacks explicit delimiters for untrusted data processing.
Capability inventory: The skill provides logic for shell command generation, API mutations, and file processing.
Sanitization: The documentation recommends using structured identifiers (NodeIDs) and least-privileged tokens to mitigate risks, though it does not specify output sanitization for processed content.

twingate-api