sliding-window-chat-truncation
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill provides a technical protocol for context management. The included Python implementation is a pure utility function that performs list slicing and dictionary formatting without interacting with the operating system, network, or external data sources.\n- [INDIRECT_PROMPT_INJECTION]: The skill involves a message ingestion surface as it processes conversation history. 1. Ingestion points: The messages parameter in the apply_sliding_window_governor function in SKILL.md receives external chat history. 2. Boundary markers: The function inserts a descriptive header [HISTORICAL MILESTONE SUMMARY: Pruned older turns for efficiency] when a summary is present to distinguish it from recent messages. 3. Capability inventory: The skill does not possess capabilities such as file writing, network access, or shell execution; it only returns a transformed list of dictionaries. 4. Sanitization: The function performs structural management rather than content sanitization, but its limited capability scope prevents the exploitation of ingested data.
Audit Metadata