edr-bypass-re
Warn
Audited by Gen Agent Trust Hub on Nov 20, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the agent to obtain external security tools from unverified personal repositories, specifically citing 'github.com/hasherezade/pe-sieve' for memory analysis and 'SysWhispers3' for syscall stub generation.
- [REMOTE_CODE_EXECUTION]: The workflow requires downloading binaries or cloning repositories from external sources and executing them (e.g., running 'pe-sieve64.exe' or 'SysWhispers3' scripts) on the target Windows environment.
- [COMMAND_EXECUTION]: The skill performs system reconnaissance using PowerShell commands like 'Get-Service' and 'fltmc filters' to identify installed security products and kernel drivers.
- [INDIRECT_PROMPT_INJECTION]:
- Ingestion points: The skill ingests and processes telemetry data, service lists, and memory dumps (e.g., hooks_dump) from the system under test.
- Boundary markers: The skill includes explicit instructions for authorized scope and isolated lab usage, along with placeholders for sensitive data.
- Capability inventory: The skill has the capability to execute shell commands, run downloaded binaries, and generate code stubs.
- Sanitization: It instructs the user to anonymize hostnames, IPs, and user information before the data leaves the lab environment.
- [DYNAMIC_EXECUTION]: The skill utilizes tools to dynamically generate customized assembly stubs (via SysWhispers3) and implements runtime patches for system components like AMSI and ETW to evaluate defensive countermeasures.
Audit Metadata