add-memory-trace-tags

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [Indirect Prompt Injection] (HIGH): User input is used to modify critical system source code without boundary markers or strict sanitization. \n
  • Ingestion points: User-provided <标签名称> (Tag Name) and <描述> (Description) parameters in the main interaction flow. \n
  • Boundary markers: Absent; user data is directly interpolated into code templates. \n
  • Capability inventory: File-write access to 11 sensitive files in developtools_profiler and third_party_musl, including system hooks and memory management headers. \n
  • Sanitization: Relies on weak natural language constraints (e.g., 'uppercase letters and underscores') which can be bypassed by malicious input designed to break out of code syntax (e.g., closing a macro and starting a new command). \n- [Command Execution] (MEDIUM): Modification of files that are subsequently compiled. \n
  • Evidence: The skill modifies .cpp, .h, and .proto files which the user is then instructed to compile. \n
  • Risk: Successful injection into these files results in arbitrary code execution on the developer's machine during the build process.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 06:58 AM