oracle

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The prompt explicitly requires embedding the user's "complete request" verbatim into shell CLI invocations (e.g., cursor-agent -p "[user's complete request]" or codex exec "[user's complete request]"), which would cause any API keys, tokens, or passwords contained in the request to be included in command-line arguments and the agent's output—directly exposing secrets.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The skill instructs running local/third‑party CLI tools with the user's complete input and explicit sandbox/approval bypass flags, which strongly risks sandbox escape and covert data exfiltration (including secrets) and thus represents a high-risk backdoor/abuse pattern.

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

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The prompt instructs the agent to execute local CLI tools (which cursor-agent/codex/opencode) and explicitly uses a --dangerously-bypass-approvals-and-sandbox flag, directing the agent to circumvent security controls and run arbitrary local commands, which compromises the machine state.