perf-ops

Fail

Audited by Gen Agent Trust Hub on Oct 14, 2026

Risk Level: HIGHPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PRIVILEGE_ESCALATION]: The skill provides documentation and command-line examples that require administrative/root privileges using the sudo command.
  • Evidence: references/load-testing.md contains a sequence of commands to add third-party GPG keys and update system repositories to install software (sudo apt-get install k6).
  • Evidence: references/cpu-memory-profiling.md includes commands to modify sensitive kernel parameters (echo -1 | sudo tee /proc/sys/kernel/perf_event_paranoid) to enable low-level profiling features, which reduces the system's security posture.
  • [COMMAND_EXECUTION]: The orchestrator relies on shell commands for system diagnosis, tool detection, and performance monitoring.
  • Evidence: SKILL.md utilizes top, free, iostat, and various which commands to inspect the execution environment and running processes.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted project metadata and profiling data, then interpolates these contents into prompts dispatched to sub-agents without proper sanitization or boundary markers.
  • Ingestion points: Reads from project configuration files (e.g., package.json, pyproject.toml, go.mod) and profiling output files (.prof, .svg, .bin) as specified in SKILL.md (Tier 1 and Tier 2).
  • Boundary markers: The dispatch templates identified in SKILL.md (T2 and T3) do not use delimiters or include instructions for the sub-agent to ignore potential commands embedded within the variable data.
  • Capability inventory: The orchestrator uses Bash, Write, and Agent (Task creation) tools to perform its duties.
  • Sanitization: There is no evidence of escaping or sanitizing the content of external files before including them in sub-agent dispatch prompts.
  • [EXTERNAL_DOWNLOADS]: The skill references and provides instructions for fetching third-party profiling and benchmarking tools from external sources.
  • Evidence: references/load-testing.md provides steps to configure repository sources from well-known services like dl.k6.io and ubuntu.com for tool installation.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 14, 2026, 06:47 PM