agent-browser

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The prompt includes explicit examples that embed plaintext passwords, tokens, and session secrets directly into CLI commands and form fills (e.g., agent-browser fill @e2 "password123", echo "pass" | agent-browser auth save), which would require the LLM to emit secret values verbatim.

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.80). Most entries are benign placeholders or official documentation (example.com, staging/example.com, lightpanda.io, github.com/login), but the presence of an explicit "malicious.com", local endpoints (http://localhost:8080), and instructions that trigger downloads/execution without verification make this set potentially dangerous — downloads or executables from unverified domains or local/redirecting endpoints should be treated as high risk.

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

• Third-party content exposure detected (high risk: 0.90). The skill's core workflow (SKILL.md) and templates (e.g., templates/capture-workflow.sh) explicitly instruct the agent to open arbitrary URLs and ingest page content via commands like "agent-browser open ", "agent-browser snapshot -i", "agent-browser get text body", and "agent-browser eval", meaning it fetches untrusted public web content (third‑party pages) and uses that content to drive interactions and decisions.