context-optimization

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from tool outputs and user messages to perform summarization and compaction, which is a vector for indirect prompt injection.
  • Ingestion points: summarize_content and ObservationStore.mask in scripts/compaction.py take arbitrary strings as input for processing.
  • Boundary markers: The implementation lacks explicit delimiters or instructions within the summarization logic to prevent the agent from following instructions embedded in the summarized data.
  • Capability inventory: The skill does not possess high-risk capabilities; no network operations, file writing, or command execution tools were found in the provided scripts.
  • Sanitization: The summarization functions use regex to extract key terms but do not perform security sanitization on the inputs.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 05:12 AM
Security Audit — agent-trust-hub — context-optimization