AI called this Suspicious at 90.0% confidence as Dangerous Capability with low false-positive risk.
Evidence for warning
- services/agent/tools.mjs exposes bash/node/file tools to a remote agent.
- controllers/agent.mjs enables those tools only by --tools/config or interactive approval.
- services/generate/rules.mjs writes Cursor/Claude/Windsurf rule files only via qelos generate rules.
- controllers/create.mjs downloads a user-selected GitHub boilerplate and runs npm install.
Evidence against
- package.json has no preinstall/install/postinstall lifecycle hook.
- cli.mjs only registers and parses user-invoked CLI commands.
- No source evidence of covert credential harvesting or exfiltration.
- Known network requests target user-selected GitHub repos or docs.qelos.io.
Behavioral surface
SourceChildProcessCryptoEnvironmentVarsFilesystemNetworkShell
Supply chainHighEntropyStringsUrlStrings
ManifestNo manifest risk signals triggered.
scanned 88 file(s), 458 KB of source, external domains: api.example.com, api.github.com, docs.qelos.io, element-plus.org, example.com, github.com, my-app.qelos.io, pinia.vuejs.org, router.vuejs.org, vue-i18n.intlify.dev, vuejs.org, your-qelos-app.com, your-qelos-gateway.example.com