The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to download and execute shell scripts directly from the internet using bash <(curl -sSL https://higress.ai/hiclaw/install.sh), which executes remote code without prior local inspection.
[REMOTE_CODE_EXECUTION]: The Windows installation guide uses the PowerShell iex (Invoke-Expression) pattern: iex $wc.DownloadString('https://higress.ai/hiclaw/install.ps1'), which is a high-risk remote code execution vector.
[EXTERNAL_DOWNLOADS]: The skill utilizes npx in worker configurations to dynamically fetch and execute packages from the npm registry at runtime, such as @modelcontextprotocol/server-filesystem and @modelcontextprotocol/server-postgres.
[EXTERNAL_DOWNLOADS]: External dependencies and configurations are pulled from multiple non-whitelisted domains including higress.io, higress.ai, skills.sh, and nacos registries.
[COMMAND_EXECUTION]: The orchestration architecture grants agents access to powerful local and network tools, including kubectl, git, and autonomous code execution environments (hermes runtime).
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8) due to the following evidence chain:
Ingestion points: Untrusted data enters the context through collaborative Matrix rooms (SKILL.md).
Boundary markers: No explicit instructions or delimiters are provided to prevent the agent from obeying instructions embedded in Matrix messages.
Capability inventory: Agents have access to the local filesystem (server-filesystem), database tools (server-postgres), Kubernetes (kubectl), and autonomous code execution (hermes runtime).
Sanitization: There is no evidence of input validation or sanitization before external data is processed by the agents.

hiclaw-multi-agent-orchestration