package-upgrade

Pass

Audited by Gen Agent Trust Hub on Nov 4, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted external data during its package inventory and web research phases.
  • Ingestion points: Reads package.json files throughout the workspace (Phase 1) and fetches package information, changelogs, and migration guides from npm, GitHub, and official documentation (Phase 2).
  • Boundary markers: The skill includes requirements for "anti-hallucination protocols", "traced proof", and confidence scores, which serve as internal checks, but does not specify data delimiters for external content.
  • Capability inventory: The skill uses file system read access, file system write access (to tmp/analysis and tmp/reports), and web research capabilities.
  • Sanitization: No specific sanitization or validation logic is defined for the data extracted from manifests or web search results.
  • [COMMAND_EXECUTION]: The skill contains instructions that attempt to grant authorization for high-privilege tool usage.
  • Evidence: The "Codex compatibility note" includes a directive stating that skill activation (including "AI-detected" activation) authorizes the use of the spawn_agent subagent tool. While not directly executed in the primary workflow, this instruction aims to influence the agent's permission boundaries regarding sub-agent creation.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 4, 2026, 10:11 AM