skills/google/mantis/mantis-summarize/Gen Agent Trust Hub

mantis-summarize

Pass

Audited by Gen Agent Trust Hub on Jul 1, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • Data Ingestion Considerations: The skill processes local source files and 'historical_learnings.jsonl' to generate summaries. As this involves interpreting external data, it presents a surface for indirect prompt injection.
  • Ingestion Points: The traversal script reads all source files and a specific JSONL history file.
  • Boundary Markers: The skill instructions do not explicitly require the use of delimiters to separate untrusted file content from the summarization prompt.
  • Capability Inventory: The tool has the capability to write files ('mantis_summary.md') and execute scripts locally.
  • Sanitization: No explicit sanitization or filtering of the input source code is defined before it is processed by the model.
  • Dynamic Script Execution: The skill involves the dynamic generation and execution of a traversal script (Python or bash) to navigate the repository tree. This is a standard and expected mechanism for mapping codebase architectures.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 1, 2026, 07:02 PM
Security Audit — agent-trust-hub — mantis-summarize