prepareNextTurn hook is ignored - model never swaps

[Fletu94]

Summary

The prepareNextTurn hook in Mods returns successfully but the model swap is never applied. The session model stays the same for all turns, even when passing full official provider model IDs (e.g. minimaxai/minimax-m2.5).

Expected Behavior

When prepareNextTurn returns { model: "minimaxai/minimax-m2.5" }, the next turn should swap and execute with minimaxai/minimax-m2.5.

Actual Behavior

The model never changes. My usage dashboard shows deepseek/deepseek-v4-flash (the session model) used for all turns, even though the hook returns a different model ID.

Steps to reproduce the issue

cmd.hooks({
  onStop: async () => {
    return {
      continue: true,
      reason: "next prompt",
      prepareNextTurn: {
        model: "minimaxai/minimax-m2.5", // full official provider model ID
      },
    };
  },
});

Command Code Version

1.3.0

Operating System

macOS

Terminal/IDE

ghostty

Shell

command

Additional context

ec68a19a-4eaf-4ee8-a987-17e7ee7aabda session

[ahmadawais]

@vipulgupta2048

[dthyresson]

I wanted to learn more about mods and hooks, so I asked to build something that switches models per turn and found this to work:

import type { ModApi } from '@commandcode/harness';

const MODELS: readonly string[] = [
  'deepseek/deepseek-v4-flash',
  'inclusionai/ling-3.0-flash-free',
  'poolside/laguna-s-2.1-free',
  'MiniMaxAI/MiniMax-M2.5',
  'deepseek/deepseek-v4-pro',
];

export default function (cmd: ModApi): void {
  let currentIndex = -1; // -1 = user's default model active first turn
  let hasShownLoadNotice = false;

  // Show load notice on the first run_start — by then the TUI feed is definitely wired
  cmd.on('run_start', () => {
    if (hasShownLoadNotice) return;
    hasShownLoadNotice = true;

    if (MODELS.length === 0) {
      cmd.ui.notify('⚠️ Model Cycler loaded but model list is empty — no cycling will occur');
    } else {
      cmd.ui.notify(`🔄 Model Cycler loaded (${MODELS.length} models in rotation)`);
    }
  });

  cmd.hooks({
    prepareNextTurn: ({ turnNumber }) => {
      if (MODELS.length === 0) return undefined;

      currentIndex = (currentIndex + 1) % MODELS.length;
      const nextModel = MODELS[currentIndex];

      try {
        cmd.ui.notify(`🔄 Model Cycler → ${nextModel} (after turn ${turnNumber})`);
      } catch (err) {
        cmd.ui.notify(
          `⚠️ Model Cycler failed to switch: ${err instanceof Error ? err.message : String(err)}`,
        );
        return undefined;
      }

      return { model: nextModel };
    },
  });
}

Few things:

• I know it worked because on switching to the free laguna or ling models I got connection issues :)
• I think that because:

Fires at the bottom of a continuing round. Results are merged across mods - later mods override earlier ones per field. A returned model/effort applies from the next model call onward (config is never mutated).

config is not changed you cannot simply use /model after and see it change because it didn't modify config.

But it def used a different model as seen in my usage:

[dthyresson]

Steps to reproduce the issue

cmd.hooks({
onStop: async () => {
return {
continue: true,
reason: "next prompt",
prepareNextTurn: {
model: "minimaxai/minimax-m2.5", // full official provider model ID
},
};
},
});

C

Also, looking at the onStop docs I don't see it allowing a prepareNextTurn param:

onStop({state, stopReason, turnNumber, lastAssistantText}, ctx?) → {continue?, reason?} | undefined

[dthyresson]

Can do this:

import type { ModApi } from '@commandcode/harness';

const MODELS = [
  'deepseek/deepseek-v4-flash',
  'inclusionai/ling-3.0-flash-free',
  'poolside/laguna-s-2.1-free',
  'MiniMaxAI/MiniMax-M2.5',
  'deepseek/deepseek-v4-pro',
] as const;

export default function (cmd: ModApi): void {
  let currentIndex = 0;

  cmd.hooks({
    onSessionStart: () => {
      cmd.ui.notify(`Model cycler — current: ${MODELS[currentIndex]}`);
    },

    onStop: () => {
      const from = MODELS[currentIndex];
      currentIndex = (currentIndex + 1) % MODELS.length;
      const to = MODELS[currentIndex];

      cmd.ui.notify(`Model cycler: ${from} → ${to}`);
      cmd.setModel(to);
    },
  });
}

Which switches, too: