fix(acp): route shell commands through terminal args

[hanhan3344]

Summary

• fix ACP Shell terminal execution to pass the shell executable in command and the shell invocation in args
• adapt the ACP terminal integration to the current ACP SDK response shape using terminal_id
• add a regression test covering bash and PowerShell command/args routing

Problem

When kimi acp was used behind ACP clients with terminal support, Shell commands like echo test could fail with errors like:

• spawn echo test ENOENT
• spawn ls -la ENOENT

The ACP terminal adapter passed the entire shell command string as terminal/create.command, which made clients try to spawn the full string as an executable path.

The adapter also assumed create_terminal() returned a TerminalHandle, but the current ACP SDK returns a response containing terminal_id, with follow-up terminal operations happening through separate RPC methods.

Fix

• use the existing Shell tool's _shell_args() to split shell execution into executable + args
• call terminal/create with:
  • command=<shell executable>
  • args=[shell flags..., user command]
• use terminal_id with terminal/wait_for_exit, terminal/output, terminal/kill, and terminal/release

Validation

• uv run pytest tests/acp/test_protocol_v1.py tests/acp/test_terminal_tool.py -q

---

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 2 additional findings.

[Copilot]

Pull request overview

This PR fixes ACP terminal-backed Shell execution by correctly splitting shell invocation into an executable (command) plus arguments (args), and updates the ACP terminal adapter to use terminal_id-based follow-up RPCs per the current ACP SDK.

Changes:

• Route Shell commands through terminal/create as command=<shell executable> and args=[shell flags..., user command].
• Update terminal lifecycle calls to use terminal_id (wait_for_terminal_exit, terminal_output, kill_terminal, release_terminal).
• Add a regression test covering bash and PowerShell command/args routing.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
tests/acp/test_terminal_tool.py	Adds regression coverage asserting correct command/args routing for bash and PowerShell in ACP terminal mode.
src/kimi_cli/acp/tools.py	Updates ACP Terminal tool to construct shell argv properly and use terminal_id-based ACP RPC methods.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Terminal calls self._shell_tool._shell_args(...), which is a private helper on Shell (leading underscore) and creates a cross-module dependency on an implementation detail. To make this more stable, consider exposing a public method (e.g., shell_args()/get_shell_invocation()), or moving the argv-building logic to a shared utility that both tools use, and have Terminal call that instead.

[hanhan3344]

Addressed Copilot's maintainability suggestion in follow-up commit e1d00c7:

• exposed a public Shell.shell_args() helper
• updated the ACP terminal adapter to call the public API instead of the private _shell_args() method
• added a focused unit test for Shell.shell_args()

Validation:
uv run pytest tests/acp/test_protocol_v1.py tests/acp/test_terminal_tool.py tests/tools/test_shell_invocation.py -q