feat: add /model slash command to CLI for session model switching

[baxen]

Summary

Add a /model command to the CLI that lets users view or switch the active model within a session without restarting.

Usage

• /model — displays the current model and provider
• /model <name> — switches to the given model, keeping the same provider

Details

• Carries over temperature and toolshim settings when switching models
• Guard rails for ACP providers and providers that manage their own context (e.g. Responses API)
• No-op when already using the requested model
• Tab-completion stub for future model name completion
• Reuses build_switched_model_config helper to construct the new ModelConfig with canonical limits for the target model

Bug fix

Fixed ordering of starts_with checks in tab completion dispatch — /model is now checked before /mode so that typing /model doesn't incorrectly fall into mode-flag completion (since "/model".starts_with("/mode") is true).

Files changed

• crates/goose-cli/src/session/completion.rs — /model in command list, completion stub, fixed dispatch ordering
• crates/goose-cli/src/session/input.rs — parse /model command, help text, tests
• crates/goose-cli/src/session/mod.rs — handle_model, build_switched_model_config, tests
• crates/goose-cli/Cargo.toml — added env-lock dev dependency for test

Fixes #8863

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 0b69181abd

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Route /model completion before /mode prefix match

Typing /model ... currently never reaches complete_model_names because the preceding line.starts_with("/mode") condition also matches /model, so tab completion falls into mode-flag completion instead. In practice, users entering /model get /mode suggestions, which makes the new command's completion path unreachable and breaks expected CLI behavior for this feature.

Useful? React with 👍 / 👎.

[baxen]

@builderbot review

[kyledef]

Just missing a version bump

[github-actions]

This pull request has been automatically marked as stale because it has not had recent activity for 23 days.

What happens next?

• If no further activity occurs, this PR will be automatically closed in 7 days
• To keep this PR active, simply add a comment, push new commits, or add the keep-open label
• If you believe this PR was marked as stale in error, please comment and we'll review it

Thank you for your contribution! 🚀

[DOsinga]

Nice work on this — the feature itself looks solid and the guard rails are well thought out.

One architectural request: we have a central slash command dispatcher in crates/goose/src/agents/execute_commands.rs where all shared commands live (/compact, /clear, /prompts, /prompt, /skills, /doctor). Could you move the model-switching business logic there?

Specifically:

1. Add "model" to the COMMANDS array and the match in execute_command()
2. Add a handle_model_command(&self, params, session_id) method on Agent that does the current model display (no args) and the provider-switch flow (with args) — including the ACP / manages_own_context guards
3. Move build_switched_model_config into the goose crate (e.g. alongside the handler or in model.rs)
4. Keep the CLI-specific bits (tab completion, InputResult::Model parsing) in goose-cli, but have them delegate to agent.execute_command("/model ...", session_id) or call the new method directly

This way both CLI and Desktop get the /model command through the shared path, and the Desktop won't need a separate REST-only flow for the same operation.

(ref: #8863)

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5b90d75c8f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Compare effective model settings before no-op return

The no-op guard only compares the raw requested string to current_model_config.model_name, but ModelConfig::new normalizes reasoning-effort suffixes (e.g. gpt-5.4-high becomes model gpt-5.4 with thinking_effort in request_params). This means a session currently on gpt-5.4-high will be treated as “already using” gpt-5.4, so /model gpt-5.4 cannot clear or change effort settings even though the effective config should change. Please base the no-op decision on the built target config (or include relevant params like thinking effort), not just model_name.

Useful? React with 👍 / 👎.