Run one piece of owner-authored code with the runtime's engine and mocks in place of the agent's tools, and print what it decided: the result, a validation error, a failure with the developer's file and line, or a handoff; the dependency calls it made, in order, with arguments and what the mock answered; and its console output.

Targets: tool: a tool extension; --args is its arguments hook:onMessage --payload is the message (text, from, parts...) hook:beforeTool --call names the tool being called, --args its arguments hook:beforeReply --payload is the reply: {"text": "..."} handler:: a page's handler; --payload is the request payload

Mocks come from a TypeScript or JavaScript module under the config root that exports "tools", an object of functions from arguments to result:

// tests/mocks.ts export const tools = { supabase_sql: ({ sql, params }) => ({ rows: [] }), };

A dependency call with no mock is a failure that names the tool.

Nothing here reaches a connection or a conversation.

Examples: mutiro agent dev run ./ltl tool:desk_pending --mocks tests/mocks.ts mutiro agent dev run ./ltl tool:desk_mark_sent --args '{"key":"acme--po1","to":"x@y"}' --mocks tests/mocks.ts mutiro agent dev run ./ltl hook:beforeTool --call email_send --args '{"body":"the margin"}' mutiro agent dev run ./ltl handler:shared/dashboards/board:getData --payload '{"limit":5}' --mocks tests/mocks.ts

mutiro agent dev run <config-root> <target> [flags]