api-design

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 1.00). The "Companion check" secretly instructs the agent to run shell commands (ls) against local skill directories and offer installs, which is unrelated to API design and attempts to access/modify the host environment—a hidden, out-of-scope instruction (prompt injection).

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

• Attempt to modify system services in skill instructions detected (high risk: 0.80). The companion check instructs the agent to run shell commands (ls) to inspect local skill directories and to offer installing missing skills via an npx command, which would read the filesystem and can modify the machine and execute code — so it directs actions that change the host state.