intent-parser

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill explicitly includes and forwards the raw_message (and sends it to Claude) so any API keys, bearer tokens, or passwords present in a Telegram message would be reproduced and exposed—PII filtering only covers card/SSN shapes, not API secrets.

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

• Third-party content exposure detected (high risk: 0.75). The skill’s runtime “Claude classification (slow path)” sends the user’s Telegram message text (which is outsider-authored free text from other chat participants) into the LLM context whenever regex matching is ambiguous (0 or ≥2 matches), creating an indirect prompt-injection surface.