terminal-execution

Terminal Execution Guidelines

This skill defines the standard workflows for executing terminal commands to ensure reliability and visibility of outputs, particularly in complex environments like WSL.

1. Primary Execution Method

Always prioritize structured execution tools over raw terminal input when possible.

1. Use execute_command: Call the execute_command tool from the terminal-runner MCP server. This provides the most reliable way to get exit codes and standard output programmatically.
2. Fallback: If terminal-runner is unavailable, use the standard run_in_terminal tool.

2. Handling Missing Output (WSL/Complex Environments)

In some environments (like WSL), terminal output may not be returned to the agent context correctly, or it may be truncated. If you suspect output is missing:

File-Based Capture Workflow

1. Prepare Directory: Ensure .tmp/copilot/ exists.
2. Redirect Output: Run the command redirecting both stdout and stderr to a file.
3. Read Result: Use file reading tools to inspect the output.

Common Patterns

Capture stdout and stderr:

some-command > .tmp/copilot/output.txt 2>&1

Capture and view live (if interactive):

some-command 2>&1 | tee .tmp/copilot/output.txt

Capture exit code separately:

some-command > .tmp/copilot/output.txt 2>&1; echo $? > .tmp/copilot/exit-code.txt

3. Best Practices

• File Naming: Use descriptive names for capture files (e.g., .tmp/copilot/build-log.txt, .tmp/copilot/test-results.txt).
• Cleanup: Periodically clean up the .tmp/copilot/ directory to avoid confusion with stale data.