Qa Patrol

Automated QA testing for web apps using local browser automation. Runs entirely on your machine — no data leaves, no cloud services, no external servers. Lev...

This skill appears to be what it claims: a local QA tool with optional static scans and DB checks. Before installing or running it: (1) Only provide test account credentials and point DATABASE_URL to a non-production/test database. (2) Expect the tool to navigate to the target URL and external services used by your app (e.g., Stripe checkout) — so "nothing leaves" is only true if your target and DB are local/test. (3) Level 3 static analysis will read local repo_path files, so run those scans only in repos you intend to scan. (4) Because this is instruction-only, its behavior depends on the platform's built-in browser/read capabilities — verify you trust the runtime environment. If you need extra assurance, run the skill in an isolated environment (VM/container) and review/edit the provided templates to remove or replace anything you don't want exercised.

Type: OpenClaw Skill Name: qa-patrol Version: 1.0.3 The skill is classified as suspicious due to its declared high-risk capabilities, despite explicit documentation of benign intent. It requires `read` permission to scan local files (`repo_path: ./src`) for static analysis, can connect to and execute arbitrary SQL queries against a user-provided database (`DATABASE_URL`), and can make arbitrary HTTP requests (`type: api_check` in `payments-stripe.yaml`). While the `SKILL.md` and `references/bug-patterns.md` files repeatedly emphasize that these features are for *detecting* issues in the user's *own codebase* (e.g., exposed API keys) and that no data leaves the machine, these capabilities are inherently powerful and could be misused by a malicious user or exploited if the agent were compromised. There is no evidence of intentional malicious behavior by the skill itself, but the broad access to local files, databases, and network makes it more than benign.