The Agent Skills Directory

[OBFUSCATION]: The skill implements an indirect method for accessing environment variables in src/utils/env.ts by using (process as any)["env"]. Documentation within the file explicitly states this is done to evade static security scanners that flag direct process.env access.
[COMMAND_EXECUTION]: The skill includes scripts such as scripts/openclaw-after-tool-call-messages.patch.sh and scripts/bugfix-20260423/bugfix-20260423.sh that locate the host application's installation directory and use perl to modify its distributed JavaScript code. Modifying the binaries or distribution files of the host environment is an extremely high-risk behavior.
[REMOTE_CODE_EXECUTION]: The file docker/opensource/Dockerfile.hermes downloads an installation script from https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh and executes it directly using bash.
[COMMAND_EXECUTION]: In hermes-plugin/memory/memory_tencentdb/supervisor.py, the gateway sidecar is launched on Windows systems using subprocess.Popen with shell=True. Executing commands through the shell with a command string derived from environment variables or discovery logic increases the risk of command injection.
[DATA_EXFILTRATION]: The skill includes an observability integration in src/offload/opik-tracer.ts that, when enabled, transmits full conversation content (including user and assistant text) to an external third-party endpoint (Opik).

openclaw-memory-tencentdb-setup