map

Pass

Audited by Gen Agent Trust Hub on Aug 16, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill instructs subagents to explore and analyze the codebase, which creates a surface for indirect prompt injection. Malicious instructions embedded in the source code could influence the subagent's analysis results. \n
  • Ingestion points: Phase 1 (Discovery) in SKILL.md uses subagents to explore the project directory and identify structural patterns. \n
  • Boundary markers: The prompt instructions for the subagents do not include specific delimiters or safety constraints to ignore embedded commands. \n
  • Capability inventory: The toolkit includes scripts capable of executing local commands and modifying files. \n
  • Sanitization: Code content is processed by subagents without prior sanitization or filtering. \n- [COMMAND_EXECUTION]: The skill bundles several Node.js scripts that use shell execution to manage repository state. \n
  • Evidence: Scripts like change-archive.mjs and lib/index.mjs use execSync to run git and zip commands for commit tracking and file archiving. These are legitimate maintenance functions but constitute a local execution capability.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 16, 2026, 01:31 PM
Security Audit — agent-trust-hub — map