tgravity-work-search

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.80). This skill explicitly asks the agent to request and accept users' Perplexity/Tavily API keys (secrets) via the conversation and store them into a local config file, so the LLM will handle secret values directly (exfiltration risk), even though it forbids embedding them in repo files or command arguments.

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

• Third-party content exposure detected (high risk: 0.75). Runtime LLM context is fed by Tavily/Perplexity API responses: tavily_search() returns data.get("answer") and perplexity_search() returns content from third-party search engines, which may include outsider-authored free text scraped from public web pages; this is then printed/used as the agent’s context via to_markdown().