openclaw/terminal-command-execution

Terminal Command Execution

Overview

Use this skill to run terminal commands with minimal risk and predictable outcomes. Prefer fast inspection, explicit intent checks, and verification after each state-changing step.

Workflow

1. Clarify goal and scope.

• Infer the exact command target from context (path, service, tool, environment).
• If request is ambiguous and risky, ask one concise clarifying question.

2. Pre-flight checks.

• Confirm working directory and required binaries.
• Inspect current state before changing it (for example ls, git status, process/listen state).
• Prefer non-destructive probes first.

3. Execute commands incrementally.

• Run the smallest command that advances the task.
• For multi-step tasks, validate each step before continuing.
• Use reproducible commands and avoid interactive flows when non-interactive options exist.

4. Handle failures systematically.

• Read stderr first and identify root cause class: permission, path, missing dependency, syntax, network, or runtime state.
• Apply one fix at a time, then re-run only the affected command.
• If privileged/destructive action is required, request user approval before proceeding.

5. Verify outcomes.

• Check exit status and observable state changes.
• For installs, verify with a version/health command.
• For edits, verify resulting files and behavior.

6. Report clearly.

• Summarize what ran, what changed, and current status.
• Include exact next command only when additional user action is required.

Safety Rules

• Avoid destructive commands by default (rm -rf, force resets, broad chmod/chown) unless explicitly requested.
• Never assume network, permissions, or package managers are available; test first.
• Prefer scoped operations (specific files/paths/services) over global changes.
• Keep secrets out of command output and logs.

Command Patterns

• Discovery: pwd, ls -la, rg --files, which <tool>
• Validation: <tool> --version, health/status commands, targeted smoke tests
• Diagnostics: inspect logs/errors first, then adjust one variable at a time