The Agent Skills Directory

[COMMAND_EXECUTION]: The utility function execute_cmd in opencode_utils.py uses subprocess.run with shell=True. In list_sessions.py, this is used to execute commands constructed from project paths retrieved via a local HTTP API, which represents a potential command injection vector if the local service provides malicious path data.\n- [COMMAND_EXECUTION]: The start_opencode function in opencode_utils.py launches background processes using os.setsid on Unix-like systems and DETACHED_PROCESS on Windows. This allows the opencode serve process to persist independently of the agent's lifecycle.\n- [DATA_EXFILTRATION]: The skill reads local project metadata and session message history from a service on 127.0.0.1:4096. While restricted to the local interface, this automated ingestion of external data into the agent's context creates a surface for indirect prompt injection.

opencode