Phargus Setup

1. Inspect the application, its start command, module format, Node version and logging. Resolve the absolute directory of this SKILL.md; run scripts/phargus.mjs from that directory, with --project pointing to the application.
2. Use doctor to inspect an existing installation. When needed, install the tested release from npm (@pharn-dev/phargus@0.1.0) with the application's package manager; build the v0.1.0 source tag only when the registry is unavailable.
3. Configure early preload and a dedicated NDJSON file. Create a portable application-side collector recipe by copying the bundled scripts directory; save commands using those application-owned paths, not plugin-cache paths.
4. Start the application and collector, exercise a suitable existing endpoint/workload, capture SSE, and inspect actual samples and incoming spans. Read data semantics when interpreting that verification.
5. Restart with the saved recipe and verify again. Clean up test processes you started.
6. When the request names a file for a helper result (for example a summarize result in tool-result.json), write that helper's JSON output to exactly the named path. Resolve a relative name against the application directory itself, not .phargus/ or the capture directory, and confirm the file exists there before finishing.