← ClaudeAtlas

yuureilisted

Run a coding agent (Claude Code, Codex) in an isolated, reproducible environment and read the resulting trace. Use when you must execute an agent task without touching the operator's real global agent configuration, or when you need a machine-readable record of a run.
shimpeiws/yuurei · ★ 2 · AI & Automation · score 68
Install: claude install-skill shimpeiws/yuurei
# yuurei `yuurei` builds a temporary execution cell per run, runs a coding agent inside it against an isolated configuration, and writes a runtime-agnostic `trace.json`. Use it to run a task reproducibly without modifying the operator's everyday global configuration. ## Install `npx skills add` installs this skill, not the CLI. If `yuurei` is not already on `PATH`, install it first: ```sh npm install -g yuurei ``` ## Shortest path Run these from the directory that will hold the project: ```sh yuurei doctor --json # 1. check the environment and auth yuurei init # 2. scaffold .yuurei/ (once) yuurei run hello --json # 3. execute the scaffolded task cat .yuurei/runs/<run-id>/trace.json # 4. read the full record ``` `init` writes `.yuurei/yuurei.yaml`, one profile, and one safe task. It never overwrites existing files and creates no credentials. It scaffolds a run named after the task, so `run hello` works out of the box; `run` prints the run ID and the run directory. ## Reading the output Every command accepts `--json`. With it, each status line is one JSON object on its own line (NDJSON), shaped `{"level": "...", "message": "...", ...data}`. `level` is one of `info`, `warn`, or `error`. Informational lines go to stdout; error lines go to stderr. ```text {"level":"info","message":"run <run-id> finished","exitCode":0,"signal":null,"runDir":"..."} ``` `yuurei trace show <run-id>` prints only a sum