Coding Agents

ToolSmith is useful when coding agents edit AI tool definitions, task fixtures, schemas, scoring, reports, importers, examples, or mock/provider behavior.

Codex should use AGENTS.md.

Claude Code should use CLAUDE.md. Claude Code projects can keep shared rules in AGENTS.md and import them from CLAUDE.md:

@AGENTS.md

Coding agents should run lint/eval/report after tool-related changes:

npm run dev -- lint examples/calendar-email
npm run dev -- eval examples/calendar-email
npm run dev -- report

Agents should summarize:

• score
• failure categories
• regressions
• changed tools and tasks
• imported OpenAPI operations, when import output changed
• recommendations

Future idea: an agent setup command could generate AGENTS.md, CLAUDE.md, and optional Claude command snippets.