The Agent Skills Directory

[SAFE]: The skill implements a secure-by-default configuration for the generated agents through the .claude_settings.json file. This configuration uses a strict permission model that explicitly denies dangerous commands such as sudo, curl, and wget, while limiting bash access to necessary development tools (npm, pip, git), adhering to the principle of least privilege.\n- [COMMAND_EXECUTION]: The skill facilitates the creation of a workspace where agents can execute shell commands. This capability is inherent to its purpose as a coding harness and is mitigated by the explicit security constraints provided in the generated configuration templates.\n- [SAFE]: The use of Archon MCP for state management (projects and tasks) introduces a structured, traceable method for preserving context across sessions, which is inherently more secure than relying on unmanaged local file storage or session-level persistence techniques.

autonomous-agent-harness