hardware-side-channel-advanced

Warn

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: MEDIUMPERSISTENCEPRIVILEGE_ESCALATIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PERSISTENCE]: The skill methodology in SKILL.md explicitly instructs the agent to establish persistence via configuration changes, scheduled tasks, or backdoors.
  • [PRIVILEGE_ESCALATION]: The Rowhammer implementation in payloads.md uses the sudo command to execute the hammerer tool with administrative privileges.
  • [EXTERNAL_DOWNLOADS]: The skill downloads code from an external GitHub repository (CMU-SAFARI/TRRespass) for Rowhammer testing.
  • [REMOTE_CODE_EXECUTION]: The skill clones, compiles, and executes third-party code from the TRRespass repository with elevated permissions.
  • [COMMAND_EXECUTION]: Python scripts in payloads.md utilize subprocess.run to execute system binaries like openssl for cryptographic timing analysis.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes hardware-derived trace data and external command results without sanitization or boundary markers.
  • Ingestion points: Trace data from ChipWhisperer and outputs from openssl.
  • Boundary markers: Absent.
  • Capability inventory: Bash, Write, subprocess, and pip.
  • Sanitization: Absent.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 19, 2026, 04:37 PM