mantis-architecture

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • Command Execution Patterns: The skill utilizes shell commands like mkdir, mv, and cp for workspace management and versioned archiving. These operations are restricted to the local environment and are functional requirements for maintaining project state.
  • Indirect Prompt Injection Surface: The skill ingests data from 'learnings.jsonl' and target source code. While the interpolation of this data into Markdown files represents a potential surface for indirect influence, the skill employs transactional logic and HTML snapshot markers to verify and record provenance. Its operations are confined to defined file system paths.
  • Security-Conscious File Handling: The implementation includes strict rules to prevent writing to the original code snapshot and uses temporary shadow directories for artifact generation, demonstrating a principle of least privilege in its design.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 05:44 PM
Security Audit — agent-trust-hub — mantis-architecture