Added recipe discovery / execution to ACP server.

[natelastname]

Summary

Adds ACP support for Goose slash command discovery and execution feedback.

This change introduces:

• Slash command discovery in ACP
  • Builds AvailableCommand list from configured slash_commands (global config)
  • Uses recipe metadata:
    • recipe.description → command description
    • parameters[].description → input hint (instead of hardcoded string)
• Execution feedback
  • When a slash command is invoked via ACP, emits a visible message:

    Running recipe: /<command>
    

  • This provides immediate user feedback, since recipe expansion is otherwise agent-visible only
• Design alignment
  • Only exposes commands explicitly configured as slash_commands (not all discovered recipes)
  • Keeps ACP behavior consistent with CLI and HTTP execution paths

Testing

I haven't included any actual unit tests, but I am happy to add some if this approach is accepted by the maintainers.

• Started Goose ACP server and connected via an ACP client (in this case, agent-shell in Emacs.)
• Verified:
  • Configured slash commands appear in available_commands_update
  • Input hint reflects recipe parameter description
  • Invoking /speckit.constitution:
    • Emits “Running recipe: …” message
    • Executes recipe successfully
    • Continues normal agent response

Related Issues

Relates to ACP integration work (no specific issue yet)
Discussion: N/A

Screenshots/Demos (for UX changes)

This is not a UX change but here's a screenshot.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 88955a934d

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Reuse command parser when emitting recipe-run feedback

This slash-command detection path uses split_whitespace() to extract the command token, but actual command execution in agents/execute_commands.rs splits only on a literal space (split_once(' ')). As a result, prompts like /recipe\targ or /recipe\narg will emit Running recipe: /recipe here even though the command dispatcher won’t resolve that recipe, so ACP clients get false execution feedback for commands that never run.

Useful? React with 👍 / 👎.