skills/gannonh/skills/upstream-assess/Gen Agent Trust Hub

upstream-assess

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes git via execFileSync in TypeScript scripts (scan-upstream.ts and intersection.ts) to perform repository analysis and metadata extraction.\n- [EXTERNAL_DOWNLOADS]: The skill fetches commit history and diffs from an external repository (https://github.com/pingdotgg/t3code.git) using standard git remote operations.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted commit messages and code diffs from the upstream repository, creating a potential attack surface for indirect prompt injection.\n
  • Ingestion points: Git commit subjects, author names, and patch diffs are retrieved from upstream/main by scripts/scan-upstream.ts and scripts/intersection.ts.\n
  • Boundary markers: The generated triage report is presented to the agent without specific delimiters or instructions to ignore embedded prompts in the raw git data.\n
  • Capability inventory: The skill possesses the capability to modify the local filesystem, create git commits, and execute shell commands (git checkout, vp check) based on the analyzed content.\n
  • Sanitization: The scripts do not perform sanitization or filtering on the content of the commit messages or code changes pulled from the external source.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 12:41 AM
Security Audit — agent-trust-hub — upstream-assess