package-upgrade

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFE
Full Analysis
  • Indirect Prompt Injection (LOW): The skill possesses a vulnerability surface where it ingests untrusted data from external sources and local files.
  • Ingestion points: Web research results from npm, GitHub, and official documentation (via WebSearch/WebFetch); local package.json files.
  • Boundary markers: The skill lacks explicit boundary markers or delimiters to separate untrusted data from instructions, though it does use a 'Context Anchor System' to refocus the agent.
  • Capability inventory: The skill is granted powerful tools including Bash, Write, Edit, and WebFetch to perform its tasks.
  • Sanitization: No explicit sanitization or filtering of the content retrieved from the web or local files is mentioned.
  • Command Execution (SAFE): Although the Bash tool is enabled, it is primarily intended for file discovery and codebase analysis (Grep/Glob). No malicious shell patterns or arbitrary command execution instructions were found.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:35 PM