The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to download and execute shell scripts directly from an external domain (higress.ai) using the bash <(curl ...) pattern. This allows the remote server to execute arbitrary commands on the user's host machine without verification.
[REMOTE_CODE_EXECUTION]: For Windows users, the skill suggests using PowerShell's Invoke-Expression (iex) to download and run a script from https://higress.ai/hiclaw/install.ps1. This method bypasses execution policies and runs unverified code from a remote source.
[EXTERNAL_DOWNLOADS]: The installation process involves adding and fetching resources from multiple external locations, including a remote Helm repository at higress.io and community-contributed skills from skills.sh, introducing unverifiable third-party code into the infrastructure.
[COMMAND_EXECUTION]: The skill requires administrative-level operations, such as modifying PowerShell execution policies (Set-ExecutionPolicy Bypass) and performing deep system integrations with Docker and Kubernetes that necessitate elevated privileges.
[CREDENTIALS_UNSAFE]: Multiple Helm installation examples guide users to pass sensitive information, such as llmApiKey and adminPassword, directly as command-line arguments using the --set flag. This practice exposes secrets in the system's process list and shell history.
[PROMPT_INJECTION]: The skill's architecture relies on ingesting untrusted data from Matrix rooms, GitHub repositories, and local filesystems. Because this data is processed by agents with capabilities like code execution (Hermes runtime) and filesystem access (MCP), it creates a significant surface for indirect prompt injection attacks.

hiclaw-collaborative-agent-os