The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches installation and uninstallation scripts from the Higress project's domains (higress.ai, higress.io) and official GitHub repositories.
[REMOTE_CODE_EXECUTION]: Installation instructions utilize shell patterns that download and execute scripts directly (e.g., bash <(curl ...) and PowerShell iex $wc.DownloadString(...)). These are standard deployment methods for the platform.
[COMMAND_EXECUTION]: The skill instructs the user to execute various CLI tools, including helm for package management, kubectl for Kubernetes orchestration, and the hiclaw CLI for agent lifecycle management.
[PRIVILEGE_ESCALATION]: Windows installation steps include Set-ExecutionPolicy Bypass, which is used to allow the execution of the platform's installation script in the current process.
[INDIRECT_PROMPT_INJECTION]: The skill defines a platform where agents ingest untrusted data from external sources (e.g., Matrix chat rooms, Nacos skill registries) while possessing high-privilege capabilities.
Ingestion points: External Matrix rooms for human-agent collaboration and the skills.sh registry for on-demand skill loading (documented in SKILL.md).
Boundary markers: The instructions do not specify explicit delimiters or "ignore instructions" prompts for these ingestion points.
Capability inventory: Agents are configured with access to the local filesystem, GitHub personal access tokens, and autonomous code execution runtimes like hermes (documented in SKILL.md).
Sanitization: The platform uses a Manager-Workers architecture where sensitive credentials are intended to remain in a centralized gateway, providing a architectural layer of isolation.

hiclaw-collaborative-agent-os