caveman-compress

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill instructs the LLM/CLI to call Claude to compress files while mandating that code blocks and inline code (and URLs/commands/env vars) be preserved EXACTLY, which forces the LLM to include any secret values contained there verbatim in its output/request.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). The skill intentionally sends user-selected local files to Anthropic/Claude and then overwrites the original memory file with model output, creating clear data-exfiltration and persistent prompt-injection/backdoor risk.

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

• Third-party content exposure detected (high risk: 0.75). The skill reads the user-specified file’s free-form markdown text (which may contain outsider-authored content) and embeds it into the Claude prompt via build_compress_prompt(original_text) → call_claude(...), so that text becomes LLM context at runtime.