Overview

Workflow

• 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.

• 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.

• 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.

• 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.

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

• 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
• Validation: --version, health/status commands, targeted smoke tests
• Diagnostics: inspect logs/errors first, then adjust one variable at a time