openclaw-update: Update OpenClaw and Carry Through to Operational Verification
機械翻訳 / Machine-translated
機械翻訳 / Machine-translated
Original article: openclaw-update: Update OpenClaw and Carry Through to Operational Verification (cccskills)
Skill page: openclaw-update
Key takeaways
Installing openclaw-update lets OpenClaw Gateway operators delegate update work to an agent.
The scope of delegation covers everything from updating using the procedure suited to your deployment environment, to post-restart session recovery, to confirming the running version.
Below, we walk through how update procedures are selected, how to install, example request phrases, and caveats to know before use.
When you ask for an existing Gateway update, the agent proceeds in one continuous flow from identifying the target to confirming it's running.
Update the OpenClaw Gateway on this server and carry through to operational verification.
The agent first investigates the target host, the running user, and where profiles and state are stored.
It also checks the service, the running release, and the management owner responsible for updates.
This is because the local repository or shell CLI may belong to a different installation.
Based on that, it selects the existing update procedure suited to the deployment environment.
Backups, restarts, and recovery proceed exactly as described in that procedure.
The final task is confirming that the target version or commit is actually running.
The result comes back as a short 1–3 line report, with a short SHA (a string identifying the commit) included only when necessary.
Repeated log lines and full hashes are not included.
This skill's job is to route to the correct update procedure for each environment and see the process through to completion.
It does not construct updates on its own.
It does not create new deployment mechanisms or schedulers.
Which procedure to use for the update is determined by the type of deployment environment.
| Deployment environment | Update procedure used |
|---|---|
| Official main checkout and macOS app | openclaw-live-updater skill |
| Team server configured by the operator | update-team-server skill |
| Other managed environments such as containers or declarative installs | Environment-specific skills or operational runbooks, plus the existing management owner |
| Standard package, or single-user source installation | OpenClaw update documentation and the update CLI (using the CLI for the target host and profile) |
Commands, locks, backup and migration requirements of the chosen procedure are followed as-is.
Notification policies and cleanup are as well.
The standard CLI update is never applied inside a separately-managed release or a running container.
Rewriting files in use by a running Gateway is also avoided.
Checkouts with uncommitted changes are never overwritten.
Update requests are treated as implying a normal restart and a limited-duration interruption.
Sessions subject to recovery are resumed automatically, so you will never be re-prompted for confirmation solely because an agent or task is running.
The graceful shutdown provided by the management owner (a method that waits for in-progress work to finish before stopping) is used for stopping.
Canceling tasks, interrupting sessions, and clearing queues can cause work to be lost.
Therefore, they are not used for the purpose of forcing a standby state.
Recovered conversations are not manually re-run.
Even if an update is deferred, fails, or rolls back, the skill does not abandon the request.
It tracks the situation until the management owner confirms completion or a specific blocking factor requiring external action is identified.
An old process simply coming back up does not count as completion.
If the user instructs a pause or cancellation, that instruction is followed.
If there is no path to proceed, the blocking factor is reported without making promises that cannot be kept.
The management owner's verification procedure is used for checking.
The main points examined are:
A successful build or receipt of a handoff is not evidence of completion.
Even if a monitoring wrapper returns success, the actual update may have been deferred or failed.
Therefore, the results of the two are read separately.
There are two ways to install: running a single command, or asking an agent.
When installing via command, you can choose the target agent such as Claude Code or Codex.
npx skills add https://github.com/openclaw/openclaw --skill openclaw-update
When asking an agent, send the following text as-is.
次の Agent Skill をインストールしてください。
https://github.com/openclaw/openclaw/tree/main/.agents/skills/openclaw-update
まず SKILL.md と付属ファイルの中身を確認し、不審な処理（外部への送信、認証情報の読み取りなど）がないことを確かめてから、Claude Code なら ~/.claude/skills/openclaw-update/、Codex なら ~/.agents/skills/openclaw-update/ に配置してください。
The installation location varies depending on the scope of use:
| Agent | Use across all projects | Use in a specific project only |
|---|---|---|
| Claude Code | ~/.claude/skills/ | .claude/skills/ |
| Codex | ~/.agents/skills/ | .agents/skills/ |
The included files are SKILL.md and agents/openai.yaml.
The procedures for routing destinations are in other skills in the same repository or in OpenClaw documentation.
If there are skills that apply to your environment (such as openclaw-live-updater or update-team-server), install those as well.
Requests can be written in your natural language.
Adding the scope of the update or what you want verified makes the report more specific.
Update my local OpenClaw Gateway to the latest version while keeping the current update channel.
In this case, "latest version" is treated as meaning: maintain the configured channel.
The agent confirms the management owner and updates using the applicable procedure.
The report afterward is a short one conveying the running version.
I want to update the Gateway on the server I manage.
First, look into which update procedure I should use.
The agent investigates the host, running user, service, and running release, and determines which procedure in the table above applies.
If the local CLI belongs to a different installation, that is identified as well.
Update to the latest main, then verify the running commit and health.
Specifying "latest main" explicitly selects the flow for updating from development-branch source.
This is the equivalent of --channel dev in the standard CLI.
Development-branch pre-checks may select an older buildable commit.
Therefore, the target is shown before the update, and the actually-running SHA is reported afterward.
The scope is limited to routine updates of existing Gateways.
New installations, release validation and publishing, and unrelated hosts are not handled.
Access to the managed procedures for the environment being updated is required.
Updates involve a restart.
Sessions subject to recovery are resumed automatically, but PTYs (pseudo-terminals) are not maintained.
If the management owner does not have a controlled interruption mechanism, the agent will report that gap or repair it within the permitted scope.
It does not create force flags or bypass compatibility checks.
Unauthorized downgrades or plugin permission changes are not automatically accepted.
Constraints and update locks specific to the managed environment are also respected.
If a definite bug is found in the update procedure, it is fixed within the permitted scope.
Fixes go through PR review and CI before being applied.
License is MIT.
Before installing, read SKILL.md and the accompanying files and verify there are no suspicious operations.
The skills that openclaw-update routes to may be used alongside it depending on the environment.
openclaw-live-updater is a skill that updates OpenClaw on macOS to the latest version of main and verifies that the Gateway and app are working.
In environments with an official main checkout and macOS app, this skill handles the update.
update-team-server hands off a Team server update to the responsible person and carries through to verification, recovery, and completion confirmation.
If you operate a Team server, install this alongside openclaw-update.
Yes.
Update requests imply a normal restart and a limited interruption, and sessions subject to recovery are resumed automatically.
However, PTYs (pseudo-terminals) are not maintained.
The local repository or CLI may belong to a different installation than the environment you want to update.
This skill confirms the target host's profile and management owner before selecting the procedure for that environment.
No.
The scope is limited to updating an existing Gateway; new installations and release validation or publishing are out of scope.
Even if an update is deferred, fails, or rolls back, it tracks the situation until completion is confirmed or a specific blocking factor is identified.
If external action is required, that blocking factor is reported briefly.
openclaw-update is a skill that updates a running OpenClaw Gateway following the managed procedures for its environment.
It carries through in one continuous flow from post-restart session recovery to confirming that the target version is running.
It is well-suited for operators who have multiple deployment environments and are unsure which update procedure to use.
Start by asking it to update your local Gateway and see how the routing and reporting flow works.
This article is a reprint of cccskills's piece "openclaw-update: Update OpenClaw and Carry Through to Operational Verification."
Skill details and installation instructions can be found on the cccskills openclaw-update page.