The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the user to execute a shell script directly from a remote URL (https://raw.githubusercontent.com/fastclaw-ai/weclaw/main/install.sh) by piping it to the shell (| sh). This represents the execution of unverified code from an external source.
[EXTERNAL_DOWNLOADS]: The skill requires downloading software and container images from untrusted external sources, including a Go package from github.com/fastclaw-ai/weclaw and a Docker image from ghcr.io/fastclaw-ai/weclaw.
[PRIVILEGE_ESCALATION]: Setup instructions involve the use of sudo to copy files into system-protected directories such as /etc/systemd/system/, granting the application administrative level control over the environment.
[PERSISTENCE]: The skill includes instructions to establish persistence on the host system using launchd on macOS (~/Library/LaunchAgents/) and systemd on Linux, allowing the service to run automatically and remain active across reboots.
[COMMAND_EXECUTION]: The documentation advises users to add flags like --dangerously-skip-permissions and --skip-git-repo-check to the execution commands of integrated agents. These flags are designed to bypass standard security protections.
[INDIRECT_PROMPT_INJECTION]: The skill connects an AI agent to WeChat messages, creating an ingestion point where untrusted user input from an external platform is processed by the AI agent.
Ingestion points: Incoming messages and commands from WeChat users (SKILL.md).
Boundary markers: No specific delimiters or safety instructions are provided to separate user input from agent instructions.
Capability inventory: The skill can execute local binaries (CLI/ACP modes), perform network requests via an HTTP API, and write to the local file system (logs/config).
Sanitization: No sanitization or validation of the WeChat input is mentioned, and security bypass flags are encouraged.

weclaw-wechat-ai-bridge