automatelisted

# Automate A protocol for turning a repetitive task into a reliable automation script. **Tradeoff**: bias toward scoping and hardening over speed. For a one-off you will run once and watch, use judgment. ## 1. Scope the task before writing code - What is the task, in one sentence? (Two sentences means two tasks; split them.) - How often, and what triggers it (manual, schedule, file change, event)? - Inputs and outputs, including exit code and side effects. - Edge cases: empty input, partial input, already-done state, missing dependency, no network. - Cost of running it twice, or failing halfway. ## 2. Pick the simplest tool that fits Bash for file moves and CLI glue. Python for branching logic, structured data, and API calls. Makefile for dependent multi-step graphs. A scheduler (systemd timer or cron) wraps one of these for unattended runs. If the task is three shell commands, it is a script, not a program. ## 3. Implement the smallest version that works Write the path that handles the real inputs first. No speculative flags or config for a single caller. Bash scripts open with `set -euo pipefail`. ## 4. Add error handling and idempotency Validate inputs before acting. Make a second run a no-op, not a duplicate or corruption. Fail loud, never swallowing an error to keep a green exit code. Log what it did. Guard destructive steps. ## 5. Dry-run, then make it runnable on a schedule Add a `--dry-run` that changes nothing and run it first. Test the edge cases deliber