automation: Docs update for OpenHands v1.9.0 release

[DevinVinson]

Release

• Version: v1.9.0
• Published: 2026-08-03T18:49:56Z
• Release: https://github.com/OpenHands/OpenHands/releases/tag/v1.9.0

Release summary

v1.9.0 adds manifest-driven automation setup and automation-interface support, a compact Cloud versus Agent-server backend chooser, and live in-chat agent activity. It also improves failed-message handling, automation run-status rendering, default-profile LLM readiness, runtime-service discovery, and local dev cleanup. The release includes terminology, telemetry, internal contract, documentation, and dependency maintenance.

Documentation review checklist

• Add a domain-neutral extension-manifest host (#16127) — Documentation needed. Document manifest-driven recommended-automation setup: prerequisite checks, form validation, review, direct creation, and conversation-assisted fallback.
• Compact Cloud versus Agent-server add-backend chooser (#16211) — Documentation needed. Update backend connection guidance and screenshots for the Cloud/Agent-server tabs, Cloud Advanced settings, and collapsed local/remote Agent-server guidance.
• Show live agent activity in chat (#16110) — Documentation needed. Update chat-interface guidance to explain the running activity chip, action-specific labels, Thinking fallback, and its disappearance when paused or complete.
• Mock setup contract from published fixtures (#16221) — Probably no documentation change. This makes mock/development behavior match the published automation setup contract.
• Drive automation UI from the interface manifest (#16222) — Documentation needed. Update automation documentation for manifest-owned navigation/copy/forms and the “Set up in a conversation” fallback where a direct setup is unavailable.
• Use “Default” for the Agent Profile badge (#16206) — Probably no documentation change. Audit Agent Profile screenshots and terminology; profile behavior is unchanged.
• Allow dismissing failed pending user messages (#16142) — Documentation needed. Document that failed messages offer Retry and Dismiss, and that Dismiss removes the bubble without returning text to the composer.
• Render CANCELLED and SKIPPED automation run statuses (#16198) — Documentation needed. Add CANCELLED and SKIPPED to automation-run status reference, including that SKIPPED can result from a concurrency limit.
• Do not gate the composer on the seeded default profile’s LLM reference (#16200) — Needs maintainer confirmation. Clarify only confirmed default-profile/active-LLM behavior; no settings migration is introduced.
• Route runtime services through server_info (#16090) — Needs maintainer confirmation. Review self-hosting and runtime-service documentation for split-stack behavior and supported /server_info.runtime_services contracts.
• Clean up dev services on SIGHUP (#16239) — Probably no public documentation change. Consider a contributor/development note for POSIX launcher cleanup.
• Rename frontend conversation-start telemetry (#16269) — Needs maintainer confirmation. Verify whether public telemetry or analytics documentation names the internal event; otherwise no end-user docs change.
• Fix runtime-modes architecture table (#16163) — Probably no documentation change in this repository. The source-repository architecture document now names dev, dev:minimal, and dev:static.
• Refresh AGENTS.md guidance (#16260) — Probably no documentation change. Repository contributor guidance only.
• Use canonical client error events (#16214) — Probably no documentation change. Internal type refactor with no documented API behavior change.
• Bump @openhands/extensions to 0.13.0 (#16258) — Documentation needed. Confirm and document the GitHub PR Reviewer and GitHub Repo Monitor setup paths, including their direct-mode fallback messages.

Pull request review

• #16127, #16221, #16222, #16258: The automation UI now consumes validated extension manifests. Recommended automations can check deployment capabilities and prerequisites, validate/preflight form data, create directly, or offer an assisted conversation fallback. Published 0.13.0 catalog entries include direct-mode fallback messages.
• #16211: Backend onboarding is functionally the same but its navigation and disclosure states changed materially; existing procedural docs and screenshots may be stale.
• #16110: Chat now surfaces the current unresolved tool/action as an accessible, non-interactive activity chip while the agent runs.
• #16142: A failed optimistic message can be dismissed; Retry remains unchanged and Dismiss intentionally does not repopulate the composer.
• #16198: Automation detail no longer fails for CANCELLED/SKIPPED runs and degrades unknown future statuses to a neutral badge.
• #16200: For the local seeded default agent profile, readiness follows the active LLM profile, matching launch behavior rather than a stale profile reference.
• #16090: Runtime-service metadata moves from frontend launch-time injection to active-backend /server_info; full dev/static stacks advertise it while dev:minimal deliberately does not.
• #16206, #16239, #16269, #16163, #16260, #16214: These are terminology, local-development, telemetry, source-doc, contributor-guidance, or internal-type changes; public documentation impact is limited or needs confirmation.

Human testing required

• Recommended automation setup: In a supported deployment, configure GitHub PR Reviewer and GitHub Repo Monitor. Verify capabilities/prerequisite messaging, required-field validation, review-before-create behavior, error mapping, and that an unsupported direct path offers the documented conversation-assisted setup without creating an automation prematurely.
• Backend connection flow: From Manage backends, test Cloud (including Advanced host settings and device-flow waiting state) and Agent-server Local/Remote connections. Verify the documented control locations, collapsed guidance, successful connection, and error recovery.
• Live activity chip: In a real multi-step conversation, run file reads, searches, and terminal commands. Verify labels change with unresolved work, completed actions disappear, Thinking appears when appropriate, pause removes the chip, and long labels remain usable on a narrow viewport.
• Failed-message dismissal: Induce a real send failure, then verify Retry resends as before and Dismiss removes only the failed bubble without restoring text or disrupting later messages.
• Automation statuses: Cancel an automation run and force/observe a concurrency-limited skipped run. Verify both remain visible on the detail page with correct neutral status labels; verify unknown statuses do not break the page.
• Default-profile LLM readiness: Locally create an authenticated active LLM profile and an existing keyless referenced profile on the seeded default agent profile. Verify the composer is enabled and a conversation uses the active profile; verify it remains blocked when the active profile has no usable credentials.
• Runtime-service and split-stack behavior: Test supported full-stack and dev:minimal setups plus a split frontend/backend deployment. Verify runtime-service context appears only when the selected backend advertises it through /server_info, with no stale frontend-injected data.

Suggested documentation locations

• /openhands/usage/automations and recommended-automation setup guides
• /openhands/usage/backends or backend connection/settings guidance
• /openhands/usage/agent-canvas chat interface and conversation troubleshooting guides
• Agent and LLM profile settings guidance
• Self-hosting, development-stack, and runtime-service reference areas — Needs maintainer triage
• Telemetry/privacy documentation — Needs maintainer triage

Source links

• Release: https://github.com/OpenHands/OpenHands/releases/tag/v1.9.0
• PRs: #16127, #16211, #16110, #16221, #16222, #16206, #16142, #16198, #16200, #16090, #16239, #16269, #16163, #16260, #16214, #16258

---

This issue was created by an AI agent (OpenHands) on behalf of the user.

[DevinVinson]

A draft documentation PR is ready for human review: #684

This comment was created by an AI agent (OpenHands) on behalf of the user.