---

name: cli-and-operations description: "Operate the DeepXiv command-line interface safely: choose commands and flags, preserve output streams, manage tokens and local configuration, and recover from CLI errors." disable-model-invocation: true metadata: disco-role: operating license: MIT

DeepXiv CLI and operations

Use this route when the task is about invoking deepxiv, selecting CLI flags, redirecting output, configuring credentials, or diagnosing a CLI failure. Keep shell commands and token/configuration handling here; use the linked routes for API-level research and local-agent behavior.

Operating route

1. Establish that the executable is the DeepXiv 1.0 CLI and inspect deepxiv --help / deepxiv --version before relying on a flag. The expected source command set and options are in the CLI reference.
2. Select the smallest command for the task. Use ask for a cited answer, search for candidate papers, paper/pmc/biorxiv/medrxiv for direct retrieval, and trending for a social-signal list.
3. Choose the backend-specific flags before running ask: arXiv uses --top-k; web uses --search-type, --gl, and --hl. Do not mix these partitions. See the reference.
4. For an answer that will be piped or saved, preserve the stream contract: answer text is stdout; citations, source annotations, quota notices, and progress are stderr. Use --json only when a complete structured payload is wanted.
5. Before an authenticated call, resolve credentials using configuration guidance. Never put a real token or LLM key in a transcript, committed file, shell history, process listing, or generated report.
6. On failure, classify the message using troubleshooting before retrying. In particular, a 403 from ask means the key lacks registered agentic access; it is not fixed by repeatedly retrying an auto-registered SDK token.

Command boundary

This skill covers the hosted CLI surface, including agent query and agent config only as invocation/configuration boundaries. Detailed Reader methods, research sequencing, answer-source interpretation, and hosted-agent API behavior belong to reader-and-paper-research. The local LangGraph/OpenAI-compatible agent's internals, budgets, tools, and provider troubleshooting belong to optional-local-agent.

Safe verification helper

Run scripts/cli_smoke.py from any working directory to check that the installed package imports and that the CLI group exposes the expected help and version without making a network request, reading credentials, or writing configuration. It is deliberately not a live command test.