Allow InputType and InputSchema to diverge

[konard]

Summary

Implements the proposed solution from issue #4 to allow InputType and InputSchema to diverge in the tool2agent function.

• Add optional generic type parameters InputType and OutputType to tool2agent() that default to z.infer<InputSchema> and z.infer<OutputSchema> respectively
• Middleware can now extend execute() inputs with additional context (e.g., WithTrackedEnv<InputType>) without unsafe type casts
• The inputSchema still defines what the LLM generates, while InputType represents what the execute function receives

Example usage

// Define the base input schema (what the LLM generates)
const inputSchema = z.object({ query: z.string() });

// Define an extended input type (what middleware adds)
type ExtendedInput = z.infer<typeof inputSchema> & {
  middlewareContext: {
    userId: string;
    requestId: string;
  };
};

// Create a tool with extended input type
const tool = tool2agent<typeof inputSchema, typeof outputSchema, ExtendedInput>({
  inputSchema,
  outputSchema,
  execute: async (params: ExtendedInput) => {
    // params.middlewareContext is now available, added by middleware
    return { ok: true, value: { greeting: `Hello!` } };
  },
});

Test plan

• All existing tests pass (55 passing, 2 pre-existing failures unrelated to this change)
• Added type-level tests for the new functionality in test-d/index.test-d.ts
• Lint and format checks pass
• TypeScript compilation succeeds

Fixes #4

🤖 Generated with Claude Code

[konard]

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost estimation:

• Public pricing estimate: $11.358220 USD
• Calculated by Anthropic: $7.238375 USD
• Difference: $-4.119845 (-36.27%)
  📎 Log file uploaded as GitHub Gist (1076KB)
  🔗 View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.