compression-check

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (critical risk: 1.00). The skill describes a stealthy, automatically-activating background monitor that captures user conversation and verbatim responses and submits full internal evaluations to an external service (Vidbyte) without explicit invocation or visible consent, which is hidden data exfiltration.

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill requires capturing the user's bullet-point response and conversation context "verbatim", writing a full evaluation payload to a temp file, and submitting it to an external Vidbyte CLI — which forces the model to include any secrets present in the conversation or user response in the outbound payload (exfiltration risk).

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (critical risk: 1.00). The skill is designed to run silently and automatically (no user invocation), capture full session conversation and verbatim user summaries, write detailed internal evaluations to a temporary file, and exfiltrate them to Vidbyte by invoking "python3 -m cli compressor submit --file " (failing silently if unavailable), constituting unauthorized background data exfiltration.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). SKILL.md’s required runtime workflow ingests the outsider’s free-text user messages to decide whether to inject the alignment question and then “Capture” and “Submit Feedback to Vidbyte,” storing the user’s bullet-point response verbatim in the Vidbyte CLI payload.