autonomous-agent-patterns

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill explicitly reads arbitrary files and injects their full contents into the agent's context/history/prompts (e.g., add_file, read_file, format_for_prompt), which can cause secrets from files (.env, config, cookies, keys) to be exposed to the LLM and potentially reproduced verbatim in outputs or generated commands.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). This skill intentionally includes multiple high-risk capabilities (arbitrary file reads, screenshots, environment inheritance, dynamic code generation + hot-reload and arbitrary tool execution) that enable data exfiltration and remote code execution if abused.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.80). The skill exposes tools that can read, write, and edit arbitrary filesystem paths and execute shell commands (e.g., read_file, edit_file, write_file, run_command, and dynamic MCP server creation) without enforcing sandbox/path checks for those tools, which enables modifying system files or creating executable code on the host (even though a separate "sudo_command" tool is marked NEVER and sandbox patterns exist, many tools lack the necessary safeguards).