skills/dotnet/skills/dump-collect/Gen Agent Trust Hub

dump-collect

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [Privilege Escalation]: The skill provides instructions for modifying system-level settings, such as using sudo tee to update the Linux core_pattern and using elevated PowerShell to modify the Windows Registry (HKLM). These operations require administrative privileges and are used to enable automatic crash dump collection at the OS level.
  • [Privileged Container Configuration]: For Docker and Kubernetes, the skill suggests adding the SYS_PTRACE capability and potentially using the --privileged flag. These settings allow debugging tools to attach to running processes but increase the container's privilege level and security surface. The skill correctly notes these should be used with caution.
  • [Diagnostic Command Execution]: The skill executes multiple utility commands, including strings, nm, docker exec, kubectl exec, gcore, and lldb. These tools are used to identify process runtimes and capture memory state for diagnostic purposes.
  • [External Downloads]: The workflow involves fetching diagnostic tools like dotnet-dump and procdump from well-known sources (such as official GitHub repositories and established software services). These tools are necessary for the collection process.
  • [Sensitive Data Handling]: The skill's core function is to collect crash dumps, which are full snapshots of process memory. This data may contain sensitive information like credentials, keys, or personal data. The skill includes specific instructions to avoid local analysis or opening of these files to reduce exposure risk.
  • [Indirect Prompt Injection Surface]: The skill processes user-supplied data—such as process IDs, container names, and binary paths—which are subsequently used in shell commands. This ingestion of untrusted data into privileged command contexts is a security consideration, particularly as the skill lacks explicit input sanitization or boundary markers.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 09:14 AM
Security Audit — agent-trust-hub — dump-collect