upstream-sync

Pass

Audited by Gen Agent Trust Hub on Apr 29, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses the subprocess module to run git commands such as clone, branch, checkout, and log, and to execute local adaptation scripts for repository synchronization.
  • [EXTERNAL_DOWNLOADS]: It clones remote repositories from GitHub as defined in the local registry (upstream_registry.json) to fetch updates and new features.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It ingests untrusted Markdown content from remote repositories and merges it into the local codebase without sanitization or instructions to ignore embedded agent directives. Ingestion points: scripts/sync_upstream.py (via git clone and reading content). Boundary markers: absent (markers are used for code preservation, not instruction isolation). Capability inventory: subprocess execution, file writing, network access via git. Sanitization: absent.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 29, 2026, 09:44 PM
Security Audit — agent-trust-hub — upstream-sync