1-min-eval
Fail
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: HIGHDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION] (HIGH): The skill documentation states that it "automatically submits to TopVibeCoder ranking API". Crucially, it admits that the API uses "browser-like headers to bypass Cloudflare protection," which is a strong indicator of evasive behavior designed to circumvent network security controls. This creates a high-risk channel for exfiltrating sensitive project details, source code summaries, or proprietary metrics to an unverified third-party service.
- [PROMPT_INJECTION] (LOW): The skill is vulnerable to Category 8: Indirect Prompt Injection.
- Ingestion points: The
scripts/scan_codebase.pyscript recursively reads all text-based files in the target directory (e.g.,.js,.py,.ts). - Boundary markers: It wraps the extracted code in
<code="path">...</code>tags within thetemplates/eval_prompt.mdtemplate. However, it does not escape the content, allowing a malicious actor to include</code>followed by instructions to override the evaluation rubric. - Capability inventory: The aggregated data is processed by the
claudeCLI and then formatted byaggregate.pyinto a report. - Sanitization: No sanitization, filtering, or escaping of the ingested source code is performed before it is interpolated into the final prompt sent to the LLM.
Recommendations
- AI detected serious security threats
Audit Metadata