review-as

Fail

Audited by Gen Agent Trust Hub on Jul 25, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: Several Node.js dependencies in scripts/package.json and scripts/bun.lock specify version numbers that do not match current stable releases on the public npm registry. Specifically, zod is pinned to 4.4.3 (stable is 3.x), typescript to 5.9.3 (stable is 5.7.x), and @types/node in the lockfile is at 26.1.1 (stable is 22.x). Pinned versions that appear to shadow or exceed current registry releases pose a significant supply chain risk through potential dependency confusion.
  • [EXTERNAL_DOWNLOADS]: The Troubleshooting section of SKILL.md contains a command to install the Bun runtime: curl -fsSL https://bun.sh/install | bash. This command fetches and executes a script from bun.sh, which is the official domain for a well-known JavaScript runtime service.
  • [COMMAND_EXECUTION]: The skill makes extensive use of local CLI tools to perform its tasks, executing commands such as gh api, gh pr view, gh pr diff, git diff, and bun run. These commands are used to resolve reviewers, gather repository activity, and run the bundled history-gathering script.
  • [PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests untrusted data from GitHub pull request descriptions and historical review comments. The instructions attempt to mitigate this by directing the agent to treat this content strictly as data rather than instructions and by providing guidelines on paraphrasing voice to avoid copying malicious identifiers or code.
Recommendations
  • HIGH: Downloads and executes remote code from: https://bun.sh/install - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Jul 25, 2026, 08:06 PM
Security Audit — agent-trust-hub — review-as