Please report security issues privately by opening a GitHub Security Advisory or emailing the maintainers. Do not open public issues for undisclosed vulnerabilities.
Prose-QA runs an LLM agent against your application and writes artifacts under .pqa/runs/:
- Bash command transcripts
- Agent messages
- HTML/JSON reports
Variables listed in sensitiveEnvVars (and PQA_LLM_API_KEY for cloud providers) are redacted from transcripts, verdicts, and reports. Other page content, user data visible in the browser, or values typed by the agent may still appear in artifacts.
Treat run artifacts as potentially sensitive. Do not publish CI artifacts or local .pqa/ directories without review.
- Store test credentials in environment variables or CI secrets — never in scenario markdown.
- Cached auth state under
.pqa/auth/contains browser session data equivalent to login cookies. Protect these files like passwords.
Prose-QA depends on agent-browser for browser automation. Keep dependencies updated and review upstream security advisories.