← Back to OpenClaw guideCategory 2
Programming and automation
This is the build lane: use OpenClaw to create repeatable systems, not just one-off assistant turns.
Use this whenYou want program-building, automation loops, scheduled actions, subagents, or deeper source-level customization.
Core pattern
What good OpenClaw programming usually looks like
The strongest pattern is simple: define the artifact, define the checks, run the work, verify the result, and only then preserve or automate the workflow. Most failed “programs” are really just vague requests with no artifact target and no proof step.
Prompt to program
Convert a one-off request into a repeatable input, output, and verification path.
Tool discipline
Use the smallest set of tools that proves the work actually happened.
Repair over rebuild
When a narrow fix will recover the line, do not rebuild the whole thing.
Recommended build targets: website QA scanners, link checkers, screenshot batchers, article-bundle validators, prompt-pack generators, asset usage inventories, and watchdog scripts that recover stalled work.
Automation
How to keep work moving without fake automation
Use cron when exact timing matters. Use heartbeats when periodic judgment matters. Use subagents when the work should run in a bounded child context. Use recovery loops when the line must keep moving until done.
Cron: exact-time events, reminders, precise schedules.Heartbeat: periodic checks plus action, not just status.Subagents: bounded investigations or production steps.Recovery loops: detect stalls, fix them, resume.
Advanced path
Go deeper only when the packaged route is not enough
Source builds and extension work are real tools, but they are not the first move. Reach for them when you actually need deeper customization, not just because the docs are there.
Open the deeper reference links for this category