metering
Warn
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill uses the environment variable
$POWERSUIT_OVERLAYto determine a file system path from which it loads additional instructions and configuration files at runtime. This dynamic path computation allows the agent's core instructions to be modified by the host environment context. - [INDIRECT_PROMPT_INJECTION]: The skill directs the agent to read and follow unverified external files (such as a README.md in the overlay directory) and treat them as authoritative instruction overrides. This creates a surface for indirect prompt injection as there are no sanitization steps or explicit boundary markers to prevent these external instructions from overriding the agent's intended safety constraints.
- [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to inspect the environment (
ls -1) and run local Python build scripts (python3 observability/grafana/build.py) for dashboard generation as part of its standard workflow.
Audit Metadata