Deterministic automation scripts for OpenClaw agents.
If you're paying for the most expensive reasoning model available, every token should go toward thinking that requires intelligence — strategy, accountability, pattern recognition, creative synthesis, pushing back. If a deterministic script can do it, it should.
The telephone operator connected lines — valuable work, but you don't need an AI agent for it.
Scripts for execution, agents only for judgment.
The pattern: Run a $0 script first. Collect data, filter noise, check actionability. Only spawn an agent when there's actually something worth reasoning about.
Result: 80% of runs cost $0 (script determines "all clear"). The other 20% cost the same but with much richer, pre-filtered context for the agent.
| Script | Description |
|---|---|
gmail-promo-cleanup/ |
Trash promotional emails across Gmail accounts |
discord-activity-digest/ |
Scan Discord channels, flag unanswered mentions |
morning-orchestrator/ |
Coordinate scripts, gate agent spawning on actionability |
Each script has its own directory with a README, source, and tests.
New scripts welcome. Requirements:
- Own directory with its own
README.md - Environment-driven config — no hardcoded credentials or account-specific values
- Dry-run mode — safe to test without side effects
- Structured output — JSON for machine consumption
- Tests — at minimum, unit tests that don't hit external APIs
- Zero LLM dependency — if it needs an agent, it's not a script
MIT