agent-browser

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The prompt contains multiple examples that embed plaintext secrets directly into commands and form fills (e.g., "password123", "echo "pass" | agent-browser auth save …", and AI_GATEWAY_API_KEY=gw_your_key_here"), so an LLM generating those agent-browser commands for authentication would likely need to output secret values verbatim.

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.80). Although many entries are benign (official hosts, docs, or placeholders), the list contains explicitly malicious/untrusted hosts (e.g., https://malicious.com), generic/unknown domains, GitHub repos/releases and local/staging endpoints that can be used to deliver executables — so these URLs collectively represent a potentially unsafe download surface unless each source is verified.

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

• Third-party content exposure detected (high risk: 1.00). The skill's SKILL.md and command docs explicitly instruct the agent to open arbitrary URLs and ingest page content (e.g., "agent-browser open ", "agent-browser snapshot -i", "agent-browser get text body", and the multi-page capture/scraping examples), meaning it fetches untrusted public web pages and the agent is expected to read and act on that content as part of its workflow.