hyper-mass
Pass
Audited by Gen Agent Trust Hub on Nov 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from potentially untrusted sources including subagent replies, handoff files, and external web pages, creating an attack surface for indirect prompt injection.
- Ingestion points: Data is ingested from node handoff files in the .hyper/mass/ directory and from external URLs fetched during the planning or execution phases.
- Boundary markers: The skill enforces strict SCOPE and writes boundaries for each node to limit the impact of a compromised agent, though it relies on the LLM to adhere to 'ignore instructions' warnings rather than using structural sandboxing.
- Capability inventory: The skill possesses the ability to execute arbitrary shell commands for verification, write to the filesystem, and perform Git commit operations.
- Sanitization: Instructions explicitly forbid following commands embedded in fetched pages, handoffs, or node replies, and prohibit the inclusion of sensitive credentials in logs or handoff files.
- [COMMAND_EXECUTION]: The skill orchestrates the execution of shell commands provided in node prompts for verification purposes. These commands are dynamically defined and executed in the user's environment, representing a significant capability that must be monitored.
Audit Metadata