context-window-management
Pass
Audited by Gen Agent Trust Hub on Apr 14, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill is an instructional guide for context engineering. The provided code templates for tiered context strategies and token budget allocation follow industry best practices and do not perform any dangerous operations.
- [PROMPT_INJECTION]: Evaluation of Indirect Prompt Injection surface:
- Ingestion points: The
messagesarray processed in thesmartSummarizeandprepareContextfunctions in SKILL.md. - Boundary markers: The skill demonstrates the use of boundary markers by wrapping summarized content in system-level delimiters (e.g.,
[Earlier context: ...]). - Capability inventory: The logic involves LLM completion calls (
llm.complete) but lacks capabilities for file system access, subprocess execution, or arbitrary network requests. - Sanitization: The summarization logic acts as a primary sanitization layer by rewriting and condensing external input before it is re-injected into the context.
- [EXTERNAL_DOWNLOADS]: References industry-standard and well-known libraries such as
tiktokenandLangChainfor tokenization and context management utilities.
Audit Metadata