Detection Patterns
Understanding DeepSweep detection rules
Pattern Categories
The editor checks code and agent configuration for credential exposure, suspicious instructions, risky MCP settings and application-security patterns. The loaded checks depend on the installed version and the files reviewed. Inspect the finding and its context; a matching pattern is not proof of exploitation, and no findings is not a guarantee of safety.
Credential Exposure
Patterns detecting API keys, tokens, and secrets left in code or AI assistant configuration files.
Prompt Injection
Patterns detecting attempts to manipulate AI assistant behavior through hidden instructions, jailbreak attempts, and instruction override attacks in files like .cursorrules.
MCP Security
Patterns detecting risky Model Context Protocol configurations, including dangerous tool permissions, auto-execution risks, and untrusted server connections.
Misconfiguration
Patterns detecting insecure AI assistant and workspace settings that widen your attack surface.
Data Exfiltration & Supply Chain
Patterns detecting attempts to extract sensitive data through AI assistant behavior, and supply chain risks in the components your assistant pulls in.
Patterns in the Configuration Review
Studio and the GitHub Action review supported agent configuration sources and report observed capabilities and inferred gaps. They are not the same check as the editor code-pattern review. Inspect the reviewed sources and warnings: unreadable or unsupported configuration is not evidence that an agent has no access.