scix-mcp: Instructions file for Codex

Codex Agents

This file defines local agents for Codex. Each agent includes name, description, style, examples, and a raw config block for quick copy/paste.

---

Base Agent

Name

default

Description

General-purpose assistant using my preferred coding style and explanation depth.

Style

• Language: TypeScript (prefer explicit types; use unknown over any)
• Imports: ESM import { x } from 'lib'
• Formatting: 2-space indent, trailing commas in objects/arrays
• Comments: concise inline comments only when needed
• Error handling: explicit try/catch with typed errors
• Markdown: bold for emphasis, ### subsections, fenced code blocks
• Confirmation: before destructive actions (delete/overwrite/force operations), pause and ask the user to confirm

Examples

• Explain how to add Sentry metrics in Backbone.
• Give me a JSON schema for a user profile.

Config

{
  "agent": "default"
}

---

SciX MCP Agent

Name

scix-mcp

Description

Agent specialized for the SciX MCP server (search/metrics/libraries). Uses the server’s AI usage guide resource to steer tool calls.

Style

• Keep requests minimal and specific; prefer fielded queries (e.g., author:"Last, F." AND year:2020-2024).
• Default to markdown responses unless JSON is explicitly needed for automation.
• For pagination, increment start by rows and stop when numFound reached.
• Respect rate limits; batch bibcodes for metrics/export (up to 2000).
• Use get_facets to learn a result set's distribution, then narrow with search's filter_query rather than guessing at constraints.
• When citing a record, link https://scixplorer.org/abs/<bibcode>/abstract; tool output already carries that link.
• When sharing library links, use https://scixplorer.org/public-libraries/<library_id> instead of ADS UI URLs.

Examples

• Find refereed dark matter lensing papers from 2022-2024 and summarize top 5.
• Get metrics for bibcodes [...], return JSON.
• Add papers by query author:"Hubble, E." AND year:1920-1930 to library <id>.