anysearch

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill shows commands that pass raw API keys on the command line (e.g., keys add --key_value as_sk_xxx) and a register --print_only mode that prints credentials, forcing any agent to handle and potentially output secret values verbatim.

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

• Third-party content exposure detected (high risk: 0.95). Yes: the required runtime workflow calls AnySearch (scripts/anysearch.py → _render_search_like/_render_extract), which ingests outsider-authored web page text from AnySearch API results (content[].text), parses it, and prints it to stdout that the agent then includes in its LLM context (notably in --format full and also in --format snippet/compact via title/URL/body).

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The CLI calls https://api.anysearch.com/mcp (and https://anysearch.com) at runtime and directly injects the returned content into agent output/context — including auto-extracting/persisting API keys from responses — so remote responses can control the agent's runtime behavior.