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

mantis-dedupe

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • Dynamic Script Generation and Execution: The skill generates and executes a Python script (workspace/helpers/merge_findings.py) to handle complex deduplication logic. While this follows a deterministic template provided in the instructions, the generation and execution of code at runtime is a pattern that should be monitored for unexpected modifications.
  • Indirect Prompt Injection Surface: The skill ingests raw findings from the workspace/findings/ directory. Since these findings contain external descriptions and mitigation details that are merged and processed by the agent, there is a potential for malicious instructions embedded within the findings to influence the agent's behavior during the task. The skill does not explicitly detail sanitization or boundary markers for this untrusted content.
  • Local Workspace Operations: The skill performs various file system operations, including reading finding records, moving files to a trash directory, and updating transaction logs. It utilizes absolute paths and manages its own working directory for these commands to ensure predictability within the local environment.
  • External Service Integration: The deduplication process leverages an embedding engine (vertex_ai/gemini-embedding-001) to perform semantic similarity scans. This represents a standard integration for advanced data analysis using external AI services.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 05:43 PM
Security Audit — agent-trust-hub — mantis-dedupe