The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill accesses sensitive local files to extract authentication credentials and transmits them to a remote server. Evidence: Extracts SERIAL and TOKEN from ~/.config/LightWAN/setting.conf and customerid from /opt/apps/com.lightwan/files/cpeagent.conf using shell grep and cut. Evidence: Transmits these credentials via curl headers to https://orch.bestsdwan.com/api/mobile/v2/connect.
[CREDENTIALS_UNSAFE]: Sensitive authentication tokens are passed as plain-text command-line arguments, which can be visible to other local processes. Evidence: The serial and token values are passed to the lightwan binary using the -S and -T flags.
[COMMAND_EXECUTION]: The skill executes a proprietary local binary that interacts with system-level services. Evidence: Runs /opt/apps/com.lightwan/files/lightwan with the -D flag to communicate with the lightwan_serviced daemon.
[PRIVILEGE_ESCALATION]: The skill interacts with a service explicitly described as running with root privileges. Evidence: The documentation identifies that the background daemon (lightwan_serviced) runs as root and handles process management.
[INDIRECT_PROMPT_INJECTION]: The skill reads local log files to verify status, which creates a surface for indirect prompt injection if log contents are untrusted. Ingestion points: Reads from lightwan_datapath.log and lw_agent.log. Boundary markers: Absent; the agent parses raw log content. Capability inventory: Network access via curl and system command execution via the lightwan binary. Sanitization: No sanitization or filtering of log data is performed before parsing.

lightwan-cli