independent-review
Warn
Audited by Gen Agent Trust Hub on Aug 25, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill provides automated installation commands that pipe remote scripts directly into shell interpreters (curl|sh, curl|bash, powershell|iex).
- [EXTERNAL_DOWNLOADS]: Installation scripts are fetched from external domains including chatgpt.com, antigravity.google, and ollama.com. These are recognized as trusted organizations or well-known services.
- [COMMAND_EXECUTION]: The script independent_review.sh executes local binaries and dynamically resolves the path to the codex binary by scanning VS Code extension directories. Additionally, Windows setup instructions use the -ExecutionPolicy ByPass flag to override local security policies.
- [DATA_EXFILTRATION]: The skill is designed to send repository artifacts (plans and code diffs) to third-party AI services. This is mitigated by an integrated secret-scanning step and a local-only configuration option for the Ollama reviewer.
- [PROMPT_INJECTION]: The skill implements an indirect prompt injection surface by interpolating untrusted repository data into model instructions. (1) Ingestion points: CONTENT variable from files or stdin in scripts/independent_review.sh. (2) Boundary markers: The artifact is wrapped in BEGIN and END delimiters. (3) Capability inventory: The script executes external binaries and processes output to generate reports. (4) Sanitization: The skill performs manual secret scanning but relies on external model safety filters for content.
Audit Metadata