walkthrough

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The prompt requires the agent to read and quote file/conversation/list items verbatim (e.g., array elements from a user-supplied file) and present them back, which would force the LLM to output any secrets (API keys, tokens, passwords, cookies) contained there.

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

• Third-party content exposure detected (medium risk: 0.65). The skill’s runtime workflow explicitly ingests “the conversation — findings, suggestions, tasks, options just produced” (and also parses an “inline list in the user’s message”), which are outsider-authored free-form text when they originate from other participants, and then “pins the list” by holding the full text in the agent’s LLM context for the walkthrough.