OpenClaw

[creative567145]

@robi that would be nice, but very risky if many use it.

From what I can notice, in every 2-5th update, there are modifications added that are causing breakage.

So perhaps the most stable way is for @andreasdueren to push updates occasionally (example once per 1-3 months), but perform tests on his own OpenClaw and if needed modify the package so that all instances don't break when an update is automatically pushed.

[robi]

@creative567145 said:
From what I can notice, in every 2-5th update, there are modifications added that are causing breakage.

Really? Break how?

I haven't been using it, but know millions of others are.

Odd that it hasn't been solved by now.

[creative567145]

@robi In the GitHub repository the breaking changes are listed
https://github.com/openclaw/openclaw/releases

Example (from our current version 2026.3.23 to the current one 2026.4.11):
openclaw 2026.4.5
openclaw 2026.4.2
openclaw 2026.3.31
openclaw 2026.3.28
openclaw 2026.3.24

Plus regular upgrades to the system functionality, for example how the Videos are being generated.
But this is due to the setup that each client has performed before that.
For example, to be able to generate Videos via Gemini, we had to install custom skills and scripts. So when the update is applied that doesn't have that requirement, either that functionality will break the current process, or we will just need to re-adjust the configuration.

Many people have reported that they are performing the breakage fixes just by connecting Claude Code via SSH and letting it do it's "magic".

[robi]

@creative567145 thanks for the info, yes that would be the easiest way.

Sounds like they need to manage flavors

[andreasdueren]

@creative567145 @robi What's the preference here? My original CI/CD was set up before the community apps existed so that's why there haven't been any proper package updates. I'm happy to set up the proper pipeline, but I need to know how people want me to set this up. Should I build daily? Weekly?

I'm not actually using openclaw. I had switched to nano claw early on and am now using Hermes agent. So I can't test this before pushing updates.

[creative567145]

@andreasdueren The only preference is that it's stable and doesn't break on updates.
So in this case, perhaps once per month. This way, if there is a breakage that is dependent on the admin unique setup within OpenClaw, we can solve that for each client.

But if a breakage origin is due to the Cloudron packaging that requires modifications, I don't know what to say...ideally somebody needs to test the updates before pushing them to all instances.

[jdaviescoates]

@creative567145 said:

ideally somebody needs to test the updates before pushing them to all instances.

I've not started using any Community Apps yet myself, but I'd suggest for anything important a good pattern might be to keep auto-updates turned off and to test them first in a cloned app before applying the updates to production

[andreasdueren]

@jdaviescoates maybe. But honestly I’ve been running Ente fully automatically updated with CI/CD for a year? now and never had any issues whatsoever.

[jdaviescoates]

@andreasdueren but this thread is about OpenClaw and it sounds like that often has breaking changes.

[robi]

@creative567145 @andreasdueren I would tend to agree monthly with a quick check that it's not a bad breakage release.

Otherwise might as well run it in LAMP and have it mange itself.

[andreasdueren]

@jdaviescoates said:

this thread is about OpenClaw

Obviously, but it sounded your concerns were about community apps not OpenClaw in specific. I'll set up monthly updates then

[jdaviescoates]

@andreasdueren fair enough.

I guess it was really for any app (community or otherwise) where people are worried that auto updates might break things. e.g. some people have auto-updates for Nextcloud turned off as plugins often break things.

Saying that, it's so easy to roll back to backups on Cloudron it's a relatively low risk for most people I'd suggest.

[andreasdueren]

CI/CD fires now on the 1st of each month.

[creative567145]

Hi @andreasdueren
One of the latest OpenClaw updates broke it.

When we try to run any command, it gives this error:
Error: Missing workspace template: HEARTBEAT.md (/app/code/src/agents/templates/HEARTBEAT.md). Ensure workspace templates are packaged.

We checked with another AI and it says the temporary solution is to create the file manually inside the container. BUT as we know the Cloudron system is read-only. So this needs to be fixed by the package maintainer.

In order so that it's easier for you to apply the fix, here is some info that might be useful:

Based on deep research into the OpenClaw ecosystem and Cloudron's architecture, this is a known bug caused by a recent upstream update to OpenClaw interacting with how Cloudron packages are built. 

Here is the exact root cause and the solutions to fix it.

### 🔍 The Root Cause
In OpenClaw version **2026.5.26**, the developers introduced a strict startup/interaction check that requires the default workspace template files (specifically `HEARTBEAT.md`) to be present in the runtime image. 

Upstream OpenClaw fixed this by updating their official `Dockerfile` to explicitly copy the `src/agents/templates` directory into the final image (GitHub Issue #87302). However, because you are using a **Cloudron community app**, the package maintainer (Andreas) uses a custom `Dockerfile` specific to Cloudron. That custom `Dockerfile` has not yet been updated to include this new `templates` directory. 

Because Cloudron strictly mounts the application directory (`/app/code/`) as **read-only**, you cannot simply SSH into the container and use `touch` or `mkdir` to create the missing file.

---

#### Option 2: Notify the Package Maintainer (The Permanent Fix)
The package maintainer needs to update the repository at `https://git.due.ren/andreas/openclaw-cloudron`. 

You can open an issue on their Git platform or reply to their release thread on the Cloudron Forum with this exact message:

> **Bug Report:** The OpenClaw >=2026.5.26 update requires workspace templates to be packaged, resulting in crashes looking for `/app/code/src/agents/templates/HEARTBEAT.md` (Upstream Issue #87302). Please update the Cloudron `Dockerfile` to include: 
> `COPY src/agents/templates /app/code/src/agents/templates/` 

[andreasdueren]

@creative567145 Thanks for the detailed report. Upstream made the workspace templates (HEARTBEAT.md etc.) a hard runtime requirement, and the Cloudron package's Dockerfile wasn't copying src/agents/templates into /app/code.

This is fixed in package version 2.0.4: the build now copies the full src/agents/templates directory into the image, and I've added a build-time assertion that fails the image build if HEARTBEAT.md is ever missing again. So this particular breakage can't silently ship a second time.

You should be able to upgrade your installation and it should work as is.

[creative567145]

@andreasdueren Thank you for the fast response and fix.
We performed the update.
And now OpenClaw is stuck in process: Starting

Here are the error logs, if they are useful:

Jul 19 17:37:59 OpenClaw 2026.7.1 (2d2ddc4)
Jul 19 17:37:59 I'm basically a Swiss Army knife, but with more opinions and fewer sharp edges.
Jul 19 17:37:59 2026-07-19T15:37:59Z 
Jul 19 17:38:00 => Healthcheck error: Error: connect ECONNREFUSED 172.18.17.9:8000
Jul 19 17:38:00 [openclaw] Could not start the CLI.
Jul 19 17:38:00 [openclaw] Reason: Cannot find module '@openclaw/ai/internal/runtime'
Jul 19 17:38:00 Require stack:
Jul 19 17:38:00 - /app/code/dist/errors-XbAR6hS3.js
Jul 19 17:38:00 [openclaw] Debug: set OPENCLAW_DEBUG=1 to include the stack trace.
Jul 19 17:38:00 [openclaw] Try: openclaw doctor
Jul 19 17:38:00 [openclaw] Help: openclaw --help
Jul 19 17:38:13 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 17:38:23 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 17:38:33 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 17:38:43 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 17:38:53 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 17:39:01 => Healthcheck error: Error: connect ECONNREFUSED 172.18.17.9:8000
Jul 19 17:39:01 ==> Loading environment from /app/data/.env
Jul 19 17:39:03 ==> Starting OpenClaw gateway on port 8000
Jul 19 17:39:05 2026-07-19T15:39:05Z 
Jul 19 17:39:05 OpenClaw 2026.7.1 (2d2ddc4)
Jul 19 17:39:05 Your config is valid, your assumptions are not.
Jul 19 17:39:05 2026-07-19T15:39:05Z 
Jul 19 17:39:06 [openclaw] Could not start the CLI.
Jul 19 17:39:06 [openclaw] Reason: Cannot find module '@openclaw/ai/internal/runtime'
Jul 19 17:39:06 Require stack:
Jul 19 17:39:06 - /app/code/dist/errors-XbAR6hS3.js
Jul 19 17:39:06 [openclaw] Debug: set OPENCLAW_DEBUG=1 to include the stack trace.
Jul 19 17:39:06 [openclaw] Try: openclaw doctor
Jul 19 17:39:06 [openclaw] Help: openclaw --help
Jul 19 17:39:13 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 17:39:23 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 17:39:33 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 17:39:43 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 17:39:53 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 17:40:01 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 17:40:06 ==> Loading environment from /app/data/.env
Jul 19 17:40:08 ==> Starting OpenClaw gateway on port 8000
Jul 19 17:40:10 => Healthcheck error: Error: connect ECONNREFUSED 172.18.17.9:8000
Jul 19 17:40:10 2026-07-19T15:40:10Z 
Jul 19 17:40:10 OpenClaw 2026.7.1 (2d2ddc4)
Jul 19 17:40:10 I'm not AI-powered, I'm AI-possessed. Big difference.
Jul 19 17:40:10 2026-07-19T15:40:10Z 
Jul 19 17:40:11 [openclaw] Could not start the CLI.
Jul 19 17:40:11 [openclaw] Reason: Cannot find module '@openclaw/ai/internal/runtime'
Jul 19 17:40:11 Require stack:
Jul 19 17:40:11 - /app/code/dist/errors-XbAR6hS3.js
Jul 19 17:40:11 [openclaw] Debug: set OPENCLAW_DEBUG=1 to include the stack trace.
Jul 19 17:40:11 [openclaw] Try: openclaw doctor
Jul 19 17:40:11 [openclaw] Help: openclaw --help
Jul 19 17:40:23 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 17:40:33 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 17:40:43 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 17:40:53 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 17:41:03 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 17:41:11 => Healthcheck error: Error: connect ECONNREFUSED 172.18.17.9:8000
Jul 19 17:41:11 ==> Loading environment from /app/data/.env
Jul 19 17:41:14 ==> Starting OpenClaw gateway on port 8000
Jul 19 17:41:15 2026-07-19T15:41:15Z 
Jul 19 17:41:15 OpenClaw 2026.7.1 (2d2ddc4)
Jul 19 17:41:15 Open source means you can see exactly how I judge your config.

[andreasdueren]

@creative567145

2.0.5 is publishing now with the fix: workspace symlinks are materialized into real directories in the image, and the build now asserts that the AI runtime module actually resolves before an image can ship. So this class of breakage will fail the build going forward instead of breaking your installation.

[creative567145]

@andreasdueren Thank you.
Now it's marked as: Not responding.

Here are the error logs, hope they are useful:

Jul 19 22:11:45 OpenClaw 2026.7.1 (2d2ddc4)
Jul 19 22:11:45 Give me a workspace and I'll give you fewer tabs, fewer toggles, and more oxygen.
Jul 19 22:11:48 ◇  Doctor warnings ────────────────────────────────────────────────────╮
Jul 19 22:11:48 - Skipped Memory Core short-term recall import for /app/data/.openclaw/workspace because SQLite rows already exist; left legacy source in place
Jul 19 22:11:48 20:11:48 [state-migrations] Legacy state migration warnings:
Jul 19 22:11:48 │
Jul 19 22:11:48 │                                                                      │
Jul 19 22:11:48 │                                                                      │
Jul 19 22:11:48 │    /app/data/.openclaw/workspace because SQLite rows already exist;  │
Jul 19 22:11:48 │    left legacy source in place                                       │
Jul 19 22:11:48 │  - Skipped Memory Core short-term recall import for                  │
Jul 19 22:11:48 ├──────────────────────────────────────────────────────────────────────╯
Jul 19 22:11:48 - Skipped Memory Core short-term recall import for /app/data/.openclaw/workspace because SQLite rows already exist; left legacy source in place
Jul 19 22:11:48 Run "openclaw doctor --fix" against the mounted state/config, then restart the container.
Jul 19 22:11:48 [openclaw] Could not start the CLI.
Jul 19 22:11:48 [openclaw] Debug: set OPENCLAW_DEBUG=1 to include the stack trace.
Jul 19 22:11:48 [openclaw] Help: openclaw --help
Jul 19 22:11:48 [openclaw] Reason: OpenClaw startup migrations did not complete cleanly; refusing to report the gateway ready.
Jul 19 22:11:48 [openclaw] Try: openclaw doctor
Jul 19 22:11:57 => Healthcheck error: AbortError: The operation was aborted
Jul 19 22:12:07 => Healthcheck error: AbortError: The operation was aborted
Jul 19 22:12:17 => Healthcheck error: AbortError: The operation was aborted
Jul 19 22:12:24 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 22:12:33 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 22:12:40 => Healthcheck error: Error: connect ECONNREFUSED 172.18.17.9:8000
Jul 19 22:12:40 ==> Loading environment from /app/data/.env
Jul 19 22:12:42 ==> Starting OpenClaw gateway on port 8000
Jul 19 22:12:44 2026-07-19T20:12:44Z 
Jul 19 22:12:44 2026-07-19T20:12:44Z 
Jul 19 22:12:44 OpenClaw 2026.7.1 (2d2ddc4)
Jul 19 22:12:44 WhatsApp automation without the "please accept our new privacy policy".
Jul 19 22:12:48 ◇  Doctor warnings ────────────────────────────────────────────────────╮
Jul 19 22:12:48 - Skipped Memory Core short-term recall import for /app/data/.openclaw/workspace because SQLite rows already exist; left legacy source in place
Jul 19 22:12:48 20:12:48 [state-migrations] Legacy state migration warnings:
Jul 19 22:12:48 │
Jul 19 22:12:48 │                                                                      │
Jul 19 22:12:48 │                                                                      │
Jul 19 22:12:48 │    /app/data/.openclaw/workspace because SQLite rows already exist;  │
Jul 19 22:12:48 │    left legacy source in place                                       │
Jul 19 22:12:48 │  - Skipped Memory Core short-term recall import for                  │
Jul 19 22:12:48 ├──────────────────────────────────────────────────────────────────────╯
Jul 19 22:12:48 - Skipped Memory Core short-term recall import for /app/data/.openclaw/workspace because SQLite rows already exist; left legacy source in place
Jul 19 22:12:48 Run "openclaw doctor --fix" against the mounted state/config, then restart the container.
Jul 19 22:12:48 [openclaw] Could not start the CLI.
Jul 19 22:12:48 [openclaw] Debug: set OPENCLAW_DEBUG=1 to include the stack trace.
Jul 19 22:12:48 [openclaw] Help: openclaw --help
Jul 19 22:12:48 [openclaw] Reason: OpenClaw startup migrations did not complete cleanly; refusing to report the gateway ready.
Jul 19 22:12:48 [openclaw] Try: openclaw doctor
Jul 19 22:12:57 => Healthcheck error: AbortError: The operation was aborted
Jul 19 22:13:07 => Healthcheck error: AbortError: The operation was aborted
Jul 19 22:13:17 => Healthcheck error: AbortError: The operation was aborted
Jul 19 22:13:24 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 22:13:33 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 22:13:43 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 22:13:48 ==> Loading environment from /app/data/.env
Jul 19 22:13:50 => Healthcheck error: Error: connect ECONNREFUSED 172.18.17.9:8000
Jul 19 22:13:51 ==> Starting OpenClaw gateway on port 8000
Jul 19 22:13:52 2026-07-19T20:13:52Z 
Jul 19 22:13:52 2026-07-19T20:13:52Z 
Jul 19 22:13:52 OpenClaw 2026.7.1 (2d2ddc4)
Jul 19 22:13:52 I'm the assistant your terminal demanded, not the one your sleep schedule requested.
Jul 19 22:13:56 20:13:56 [state-migrations] Legacy state migration warnings:
Jul 19 22:13:56 - Skipped Memory Core short-term recall import for /app/data/.openclaw/workspace because SQLite rows already exist; left legacy source in place
Jul 19 22:13:56 │
Jul 19 22:13:56 ◇  Doctor warnings ────────────────────────────────────────────────────╮
Jul 19 22:13:56 │                                                                      │
Jul 19 22:13:56 │  - Skipped Memory Core short-term recall import for                  │
Jul 19 22:13:56 │    /app/data/.openclaw/workspace because SQLite rows already exist;  │
Jul 19 22:13:56 │    left legacy source in place                                       │
Jul 19 22:13:56 │                                                                      │
Jul 19 22:13:56 ├──────────────────────────────────────────────────────────────────────╯
Jul 19 22:13:56 [openclaw] Could not start the CLI.
Jul 19 22:13:56 [openclaw] Reason: OpenClaw startup migrations did not complete cleanly; refusing to report the gateway ready.
Jul 19 22:13:56 - Skipped Memory Core short-term recall import for /app/data/.openclaw/workspace because SQLite rows already exist; left legacy source in place
Jul 19 22:13:56 Run "openclaw doctor --fix" against the mounted state/config, then restart the container.
Jul 19 22:13:56 [openclaw] Debug: set OPENCLAW_DEBUG=1 to include the stack trace.
Jul 19 22:13:56 [openclaw] Try: openclaw doctor
Jul 19 22:13:56 [openclaw] Help: openclaw --help
Jul 19 22:14:07 => Healthcheck error: AbortError: The operation was aborted
Jul 19 22:14:17 => Healthcheck error: AbortError: The operation was aborted
Jul 19 22:14:27 => Healthcheck error: AbortError: The operation was aborted
Jul 19 22:14:33 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 22:14:43 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 22:14:53 => Healthcheck error: Error: connect EHOSTUNREACH 172.18.17.9:8000
Jul 19 22:14:56 ==> Loading environment from /app/data/.env
Jul 19 22:14:59 ==> Starting OpenClaw gateway on port 8000
Jul 19 22:15:00 => Healthcheck error: Error: connect ECONNREFUSED 172.18.17.9:8000
Jul 19 22:15:00 2026-07-19T20:15:00Z 

[andreasdueren]

Thanks for your patience, this was a third, different layer of the same upstream release, and should be fixed in 2.0.6.

What happened: OpenClaw v2026.7.x introduced a strict startup check. After an update, the gateway runs state migrations and refuses to start if any migration finishes with a warning. Your install has older Memory Core state files (memory/.dreams/*.json) whose contents had already been migrated to the new SQLite store, so the migration skipped them with a "left legacy source in place" warning. And that recurring warning was treated as fatal on every restart, hence the permanent "Not responding".

2.0.6 now attempts the package self-heal on startup: it runs upstream's documented remedy (openclaw doctor --fix) automatically before starting the gateway, and then archives the already-migrated legacy state files to /app/data/legacy-state-archive/<timestamp>/.

Please update to 2.0.6, it should then come up healthy with no manual steps. If it still misbehaves, post fresh logs and I'll take another look.

[creative567145]

@andreasdueren Thank you very much. I confirm that OpenClaw is now running without errors.