OpenClaw Just Rewrote Its Docs From the Ground Up — Automation Hub, Image Generation, Background Tasks, and More

The Automation Problem Nobody Talked About

OpenClaw has accumulated six distinct automation mechanisms over the past year: heartbeat polling, cron jobs, background tasks, hooks, standing orders, and webhooks. Each had its own documentation page. None of them explained when to use one over another. If you wanted to schedule a recurring check, you had to read all six pages and figure out the differences yourself.

The new Automation Hub changes that. It's a single overview page that opens with an interactive Mermaid flowchart — a decision tree that walks you through a series of questions: Do you need precise scheduling? Use cron. Reacting to events? Use hooks. Persistent instructions across sessions? Standing orders. Need to track long-running work? Background tasks.

Below the flowchart sits a comparison table covering all six mechanisms at a glance: what triggers them, how they run, and what they're best suited for. The page also covers specialized automation patterns like Gmail PubSub integration, polling strategies, and auth monitoring. For the first time, OpenClaw's automation story is told in one place.

Background Tasks Get a Complete Reference

OpenClaw quietly migrated its task ledger to SQLite earlier this month. Until now, the only documentation was a few scattered mentions in the cron jobs page. That's no longer the case.

The new Background Tasks page is a 200-line reference that covers everything: what creates tasks (ACP background runs, subagent spawns, isolated cron executions, CLI gateway operations), the full lifecycle state machine (queued, running, succeeded, failed, timed out, cancelled, lost), and the difference between direct delivery and session-queued delivery.

There's a Mermaid lifecycle diagram showing the state transitions visually. The CLI reference documents five commands — openclaw tasks list, show, cancel, notify, and audit — with explanations of notification policies, audit findings, and the automatic 60-second sweeper that cleans up stale runs. The page also clarifies what doesn't create tasks: heartbeat turns, main-session cron reminders, and normal interactive chat.

Alongside the new page, the existing cron jobs and cron-vs-heartbeat docs were updated with cross-references. An earlier incorrect claim that main-session cron jobs create trackable tasks was also corrected.

Image Generation Across Four Providers, Finally Explained

OpenClaw's image_generate tool has supported multiple providers for a while, but the documentation was essentially “set an API key and ask the agent to make a picture.” The new image generation page changes that with a proper technical reference.

Four providers are now documented in detail: OpenAI (gpt-image-1), Google (gemini-3.1-flash-image-preview), fal (fal-ai/flux/dev), and MiniMax (image-01). Each has different capabilities — only Google, fal, and MiniMax support image editing with reference images, for example — and a new capability matrix compares generation limits, editing support, size controls, aspect ratios, and resolution options across all four.

The page documents all ten tool parameters and introduces the fallback chain concept: you can configure a primary model and a list of fallbacks, and OpenClaw will try them in order. The resolution order is tool parameter first, then configured primary, then fallbacks, then auto-detection. For self-hosters juggling multiple API keys and cost tiers, this is the kind of documentation that saves hours of trial and error.

“Exec Suddenly Asks for Approval” Gets Its Own Troubleshooting Section

OpenClaw recently changed how execution approval defaults work. When tools.exec.host is unset, it now defaults to auto, which resolves to sandbox if a sandbox runtime is active and gateway otherwise. Gateway and node execution now default to an allowlist security model with on-miss approval prompts. The result: agents that previously ran commands silently now pause and ask for permission.

The new troubleshooting section in the help docs walks through the diagnostic commands, explains the behavioral shift, and offers three resolution paths: full restoration of the previous behavior, a balanced middle ground that keeps approvals but adjusts routing, or switching to sandbox mode entirely. It also documents the common log messages users will encounter, so they're not left guessing why their automation stopped working.

The Docs Sidebar Got a Structural Overhaul

Beyond the new content pages, the documentation navigation itself was restructured. The “Nodes and devices” section was renamed to “Nodes and media” and split into two logical subgroups: media capabilities (media understanding, images, audio, camera, text-to-speech) and node features (talk, voice wake, location commands). Misplaced navigation entries were corrected.

Perhaps more significantly, Related sections were added to every channel documentation page and every automation page. Heartbeat links to cron. Cron links to background tasks. Agent loop links to multi-agent. Sessions link to timezone handling. The documentation is no longer a collection of isolated pages — it's starting to form a connected graph where every concept links to its neighbors.

What Shipped

Why This Matters

OpenClaw has been shipping features at a blistering pace. Background tasks, multi-provider image generation, exec approval overhauls — all of it landed in the codebase weeks before the documentation caught up. For a self-hosted project where users are expected to configure, troubleshoot, and maintain their own instances, that gap is expensive. Every undocumented feature is a support ticket waiting to happen or, worse, a user who gives up and moves on.

What makes this push notable isn't just the volume — 25 commits touching documentation in a single day — but the structural thinking behind it. The Automation Hub doesn't just list features; it helps users choose between them. The Related sections don't just link pages; they build a navigable knowledge graph. The Background Tasks page doesn't just describe the feature; it documents the CLI, the lifecycle, the audit tooling, and the edge cases.

Documentation is often the last thing open-source projects invest in. OpenClaw just spent a Sunday making its docs match its code. If they keep this up, it could become the thing that sets the project apart — not just what it can do, but how well it explains what it can do.