Text Area
Coding assistants operate in a repository and can often read files, edit code, and run commands. They are not substitutes for a scientific workflow architecture or a distributed agent runtime.
Coding Assistant Reference Cards
Text Area
| Tool | Choose it when... | Do not choose it solely because... | Project context |
|---|---|---|---|
| Claude Code | The team wants a terminal coding workflow centered on the Claude model family, uses CLAUDE.md, or wants Claude-specific planning and tool patterns | It appears first in a list; validate model availability, repository fit, and team preference | CLAUDE.md; also use AGENTS.md for portable guidance |
| Codex CLI | The team prefers Codex-style repository work, OpenAI-compatible configuration, AGENTS.md, or noninteractive/scriptable runs | An example model name appears in a card; query the current MAG model list | AGENTS.md |
| OpenCode | The team values an open-source client, provider portability, and a common interface across supported model protocols | More provider options automatically mean less setup or lower operational risk | AGENTS.md and optional MCP configuration |
Recommendation: choose one team default, test it on three representative tasks, and document the results. Suggested tasks are repository explanation, a bounded bug fix with tests, and a small scientific-data transformation with known expected output.