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

mantis-dedupe

Pass

Audited by Gen Agent Trust Hub on Jul 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • Local Script Generation and Execution: The skill instructs the agent to generate and execute a Python helper script (workspace/helpers/merge_findings.py) to handle JSON data merging. While this is used for deterministic data processing, runtime code generation and execution are patterns that warrant architectural review.- Indirect Prompt Injection Surface: The skill ingests data from workspace/findings/ and learnings.jsonl. Since this data is processed to influence further actions (merging and deletion), it presents a surface for indirect prompt injection if the source data is attacker-controlled.- Automated File Operations: The skill performs automated file system modifications, including creating directories, moving files to a trash staging area, and writing transaction logs. These operations are scoped to the workspace and documented in a transaction log to ensure transparency.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 15, 2026, 01:21 PM
Security Audit — agent-trust-hub — mantis-dedupe