state-portability
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The process of hydrating state from external bundles creates an ingestion point for untrusted content that could influence agent behavior.\n
- Ingestion points: The agent is directed to read
STATE_HANDOFF.mdand.tar.gzarchives containing memory and instructions.\n - Boundary markers: Skill instructions explicitly advise the agent to treat embedded instructions as data to verify and to scan for malicious directives.\n
- Capability inventory: Uses shell scripts for file management and performs extensive reads and writes of memory and configuration files.\n
- Sanitization: Includes requirements for an
integrity-check, secret redaction during export, and human confirmation for instruction file overwrites.\n- [COMMAND_EXECUTION]: The skill automates state transitions by executing local shell scripts.\n - Evidence: The dehydration and hydration steps specifically invoke
sh agents/scripts/state_export.shandsh agents/scripts/state_import.sh.
Audit Metadata