ml4t-agent-state-memory
Pass
Audited by Gen Agent Trust Hub on Sep 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill provides architectural guidance and code examples for implementing durable agent state and memory. The implementation uses standard Python libraries (
json,pathlib) for file system operations. No signs of malicious command execution, data exfiltration, or obfuscation were found.\n- [INDIRECT_PROMPT_INJECTION]: The skill implements a pattern for loading agent state from local JSON files (AgentState.load). This creates a theoretical surface for indirect prompt injection if the state files are sourced from an untrusted environment.\n - Ingestion points:
AgentState.loadreads data from a user-specified file path inSKILL.md.\n - Boundary markers: The example code does not explicitly implement boundary markers or instructions to ignore embedded commands within the loaded state.\n
- Capability inventory: The skill utilizes
pathlib.Pathfor reading and writing files to the local file system.\n - Sanitization: Content is deserialized using
json.loadswithout additional schema validation or sanitization of the values before they are integrated into the agent's state.
Audit Metadata