java-refactor-engram-contract

Pass

Audited by Gen Agent Trust Hub on Jun 22, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill defines a protocol for handling data from a memory system, representing a surface for indirect prompt injection. Analysis: 1. Ingestion points: Artifacts are retrieved via mem_get_observation (SKILL.md). 2. Boundary markers: The skill requires explicit validation of project and run_id and enforces strict namespace isolation for topic keys. 3. Capability inventory: Access is limited to platform-specific memory tools (mem_save, mem_search, mem_get_observation). 4. Sanitization: Instructions mandate compact evidence and explicitly forbid storing raw source code or logs.
  • [SAFE]: The skill implements strong security hygiene by preventing session-based inference and enforcing explicit decision gates before data access.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 22, 2026, 05:53 PM
Security Audit — agent-trust-hub — java-refactor-engram-contract