The package contains dangerous AI-agent execution capabilities but no lifecycle hook, self-dependency, covert payload, or confirmed credential exfiltration. It warrants a warning for capability abuse risk rather than a malware block.
Static reason
No blocking static signals were detected.; previous stored version diff introduced dangerous source
Trigger
An operator configures and starts the manager, then a managed agent invokes host shell_exec.
Impact
A compromised AWB control plane or unsafe agent instruction could execute commands and access the manager user's files and credentials.
Mechanism
MCP-mediated arbitrary host command execution with permissive Claude defaults.
Evidence
dist/lib/managed-agent-store.jsdist/lib/host-mcp/server.jsdist/lib/cli-adapters/claude.js$AWB_AGENT_MANAGER_HOME/agents/<agentId>/mcp-config*.json$AWB_AGENT_MANAGER_HOME/agents/<agentId>/cli-home/.credentials.json