narumiruna Using axiv CLI
Use only the published axiv CLI commands in this repository.
Use an installed axiv executable when available.
From this source checkout, fall back to uv run axiv as the command prefix when axiv is not installed.
Read references/installation.md when neither command prefix can show CLI help.
Stop and report the installation error when the installation instructions do not produce a working command prefix.
Do not select or retry MCP transports manually because axiv handles streamable HTTP to SSE fallback only before tool dispatch.
When an authenticated command reports an unexpected runtime, MCP connection, initialization, or breaking tool drift error, update axiv by following references/installation.md before troubleshooting.
After updating, rerun only non-quota, read-only checks such as axiv --help or axiv auth status --json.
Treat a surfaced tool-call transport error as terminal because the remote outcome may be ambiguous.
Do not automatically retry a quota-consuming command or remote library write after updating.
Do not invoke mcp2cli, call alphaXiv REST or MCP endpoints directly, or import private Python client methods.
Run the resolved command prefix with auth status --json before any authenticated research or library workflow.
Stop and report the error when the API key is missing, authentication fails, a reviewed tool is missing or changes schema, or alphaXiv returns 403.
Read references/command-map.md to select the exact command and understand its authentication, quota, and remote effects.
Prefer --json whenever another tool or agent will consume the result.
Preserve the user's question and keywords instead of silently expanding acronyms or inventing search terms.
Tell the user that discover, content, query, and code commands consume Assistant quota before calling one unless their request already clearly authorizes that research action.
Stop and report quota exhaustion instead of retrying or substituting another quota-consuming command.
Use axiv library list --json to obtain current opaque folder IDs before planning a library change.
Before every remote library write, state the exact CLI operation, folder IDs or names, and paper IDs or URLs that will change.
Obtain explicit user authorization for those exact targets before adding --yes.
Treat --yes only as CLI confirmation and never as evidence that the user authorized the write.
Do not combine separately authorized writes or broaden a target after authorization.
For folder deletion, explicitly state that folder memberships will be removed before requesting authorization.
Stop when a write target is missing, ambiguous, destructive intent is unclear, or authorization does not match the final command.
Read references/workflows.md when coordinating literature review, paper analysis, PDF evidence, code verification, or library organization.
Verify success from the command's JSON result and report partial or failed outcomes without automatically retrying writes.