The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation instructions direct users to download and execute scripts directly from a remote server (higress.ai) using bash process substitution and PowerShell's Invoke-Expression (iex).
Evidence: bash <(curl -sSL https://higress.ai/hiclaw/install.sh)
Evidence: iex $wc.DownloadString('https://higress.ai/hiclaw/install.ps1') in SKILL.md.
[PRIVILEGE_ESCALATION]: The installation guide for Windows includes commands to bypass the system's script execution policy, which reduces the security posture of the environment to facilitate the installation of unverified scripts.
Evidence: Set-ExecutionPolicy Bypass -Scope Process -Force in SKILL.md.
[COMMAND_EXECUTION]: The skill facilitates the execution of arbitrary commands via Stdio MCP integrations, specifically using npx to run packages with access to the host filesystem.
Evidence: command: npx with args @modelcontextprotocol/server-filesystem and /workspace in worker definitions.
[INDIRECT_PROMPT_INJECTION]: The skill defines a platform for multi-agent collaboration that ingests and processes untrusted data from Matrix rooms and external MCP servers, creating a significant attack surface.
Ingestion points: Matrix chat messages, Git repositories (via MCP), and shared MinIO filesystem data.
Boundary markers: None specified in the resource definitions to delimit instructions from data.
Capability inventory: The agents have access to Kubernetes resource management (kubectl), filesystem read/write, and various LLM runtimes.
Sanitization: No evidence of input validation or prompt sanitization is provided in the configuration patterns.

agentteams-hiclaw-multi-agent-orchestration