Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

On ghcr.io/openclaw/openclaw:2026.4.22, bundled plugins that install runtime deps at gateway startup fail with npm error code EUNSUPPORTEDPROTOCOL — Unsupported URL Type "workspace:": workspace:*; 18 plugins failed in our deployment (acpx, amazon-bedrock, amazon-bedrock-mantle, anthropic, browser, deepgram, elevenlabs, fireworks, github-copilot, google, kimi, lobster, microsoft, mistral, ollama, openai, telegram, xai) and the gateway ends up ready with only the 3 dep-free plugins loaded.

Steps to reproduce

Observed on a Debian 12 VM with existing v2026.4.15 state. I have NOT tested whether a fresh install reproduces.

Precondition: existing working v2026.4.15 deployment at host path /home/simone/.openclaw, bind-mounted into the container at /home/node/.openclaw.

1. docker pull ghcr.io/openclaw/openclaw:2026.4.22
2. docker stop openclaw && docker rm openclaw

3. docker create --name openclaw --restart unless-stopped \
     -p 18789:18789 \
     -v /home/simone/.openclaw:/home/node/.openclaw \
     --env-file /home/simone/.openclaw/env-backup-pre-upgrade.txt \
     ghcr.io/openclaw/openclaw:2026.4.22
   

4. docker start openclaw
5. After ~45s: docker logs openclaw --tail 100

Expected behavior

Bundled plugins load and register; Telegram channels start their providers; the gateway ready line reports the full plugin set observed on v2026.4.15 (7 plugins in our config: acpx, browser, device-pair, lobster, phone-control, talk-voice, telegram). Rolling back to v2026.4.15 on the same host, same bind-mounted ~/.openclaw, and same env file produces exactly that state.

Actual behavior

[plugins] 4 plugin(s) failed to initialize (validation: acpx, browser, lobster, telegram). Run 'openclaw plugins list' for details.
[gateway] [plugins] failed to install bundled runtime deps: Error: npm error code EUNSUPPORTEDPROTOCOL
npm error Unsupported URL Type "workspace:": workspace:*
npm error A complete log of this run can be found in: /home/node/.npm/_logs/2026-04-23T19_19_37_778Z-debug-0.log (plugin=acpx, source=/app/extensions/acpx/index.ts)

Same error with distinct source paths for browser (/app/extensions/browser/index.ts), lobster (/app/extensions/lobster/index.ts), and telegram (/app/extensions/telegram/index.ts). Gateway then logs:

[gateway] ready (3 plugins: device-pair, phone-control, talk-voice; 12.5s)

~15 minutes after startup a second validation pass expanded the failing-plugin set to 18:

[plugins] 18 plugin(s) failed to initialize (validation: acpx, amazon-bedrock, amazon-bedrock-mantle, anthropic, browser, deepgram, elevenlabs, fireworks, github-copilot, google, kimi, lobster, microsoft, mistral, ollama, openai, telegram, xai). Run 'openclaw plugins list' for details.

All four Telegram bot accounts remained offline until rollback. I did not capture the contents of the referenced /home/node/.npm/_logs/2026-04-23T19_19_37_778Z-debug-0.log before rolling back. I can reproduce and capture the log on request.

OpenClaw version

2026.4.22 (ghcr.io/openclaw/openclaw:2026.4.22)

Operating system

Debian 12 (GCE e2-medium, Linux 6.1); container image is the official Linux build.

Install method

Docker, official GHCR image.

Model

minimax/MiniMax-M2.7 (primary, per openclaw.json; not reached because no model-provider plugin loaded)

Provider / routing chain

openclaw (local gateway mode) → minimax (direct); fallback openai (also blocked by plugin failure)

Additional provider/model setup details

Local gateway mode (gateway.mode: local). Bundled plugins enabled in plugins.entries: telegram, acpx, browser, lobster. Provider plugins referenced via models.providers.*: minimax, openai, anthropic. No custom registries or proxies. Container env file contains 17 vars including ANTHROPIC_API_KEY, MINIMAX_API_KEY, OPENAI_API_KEY, TELEGRAM_BOT_TOKEN.

Logs, screenshots, and evidence

2026-04-23T19:19:33.897+00:00 [gateway] loading configuration…
2026-04-23T19:19:33.981+00:00 [gateway] resolving authentication…
2026-04-23T19:19:33.983+00:00 [gateway] starting...
2026-04-23T19:19:44.731+00:00 [plugins] 4 plugin(s) failed to initialize (validation: acpx, browser, lobster, telegram). Run 'openclaw plugins list' for details.
2026-04-23T19:19:44.735+00:00 [gateway] [plugins] failed to install bundled runtime deps: Error: npm error code EUNSUPPORTEDPROTOCOL
npm error Unsupported URL Type "workspace:": workspace:*
npm error A complete log of this run can be found in: /home/node/.npm/_logs/2026-04-23T19_19_37_778Z-debug-0.log (plugin=acpx, source=/app/extensions/acpx/index.ts)
2026-04-23T19:19:44.736+00:00 [gateway] [plugins] failed to install bundled runtime deps: Error: npm error code EUNSUPPORTEDPROTOCOL
npm error Unsupported URL Type "workspace:": workspace:*
npm error A complete log of this run can be found in: /home/node/.npm/_logs/2026-04-23T19_19_38_298Z-debug-0.log (plugin=browser, source=/app/extensions/browser/index.ts)
2026-04-23T19:19:44.737+00:00 [gateway] [plugins] failed to install bundled runtime deps: Error: npm error code EUNSUPPORTEDPROTOCOL
npm error Unsupported URL Type "workspace:": workspace:*
npm error A complete log of this run can be found in: /home/node/.npm/_logs/2026-04-23T19_19_42_811Z-debug-0.log (plugin=lobster, source=/app/extensions/lobster/index.ts)
2026-04-23T19:19:44.738+00:00 [gateway] [plugins] failed to install bundled runtime deps: Error: npm error code EUNSUPPORTEDPROTOCOL
npm error Unsupported URL Type "workspace:": workspace:*
npm error A complete log of this run can be found in: /home/node/.npm/_logs/2026-04-23T19_19_44_212Z-debug-0.log (plugin=telegram, source=/app/extensions/telegram/index.ts)
2026-04-23T19:19:46.506+00:00 [gateway] agent model: minimax/MiniMax-M2.7
2026-04-23T19:19:46.508+00:00 [gateway] ready (3 plugins: device-pair, phone-control, talk-voice; 12.5s)
2026-04-23T19:36:29.886+00:00 [plugins] 18 plugin(s) failed to initialize (validation: acpx, amazon-bedrock, amazon-bedrock-mantle, anthropic, browser, deepgram, elevenlabs, fireworks, github-copilot, google, kimi, lobster, microsoft, mistral, ollama, openai, telegram, xai). Run 'openclaw plugins list' for details.

Impact and severity

• Affected: any ghcr.io/openclaw/openclaw:2026.4.22 deployment that depends on bundled provider or channel plugins (observed with the Telegram, MiniMax, OpenAI, Anthropic plugins — effectively any production deployment of the Docker image).
• Severity: critical — complete agent outage. No model provider loaded, no channel connected.
• Frequency: 1/1 observed (deterministic on first boot of v2026.4.22 container with the pre-existing state described above). I did not test whether a fresh state reproduces.
• Consequence: all four Telegram agents offline until rollback to v2026.4.15 (performed via a pre-upgrade docker commit openclaw openclaw-rollback:v2026.4.15 image; restore took ~2 minutes).

Additional information

• Last known good: v2026.4.15 on the same host and the same bind-mounted ~/.openclaw and same env file. Rolling back to v2026.4.15 via the docker commit-tagged image restored all 7 plugins and all 4 Telegram providers ([telegram] [betty][default][jade][jessica] starting provider lines present in logs).
• First known bad: v2026.4.22. I did not test v2026.4.20 or v2026.4.21 in isolation.
• Related open reports (same upgrade window, distinct install methods or channels): v2026.4.22: Telegram channels crash on startup — Cannot find package 'openclaw' in plugin-runtime-deps #70701 (Telegram, npm global, macOS), 4.22: Discord provider crash-loops with 'Cannot find package openclaw' on npm global install (ESM subpath import fails from plugin-runtime-deps cache) #70756 (Discord, npm global), [Bug]: Telegram channel keeps loading stale runtime hash after update/reinstall (Cannot find package 'openclaw' from plugin-runtime-deps) #70773 (Telegram, runtime-hash path, Linux npm global), [Bug]: Discord provider fails with "Cannot find package 'openclaw'" in profile-specific plugin-runtime-deps after global npm upgrade #70818 (Discord, profile-specific runtime-deps), [Bug]: [email protected] runtime dep not staged on npm global install of 2026.4.22 — ACP plugin fails to load with 'Cannot find module acpx/runtime' #70839 (acpx, npm global). This report documents the same window on the official Docker image and captures the install-step EUNSUPPORTEDPROTOCOL error, which is not present in those reports — they show the downstream Cannot find package 'openclaw' in plugin-runtime-deps resolve-step symptom. I have not verified whether those are the same underlying defect.
• Secondary observation (may or may not be related): v2026.4.20 introduced feat(cron): split jobs.json into config and runtime state files #63105 which splits cron runtime state out of cron/jobs.json into cron/jobs-state.json. On the affected v2026.4.22 container, ~15 minutes after startup, ~/.openclaw/cron/jobs.json still contained state fields and no cron/jobs-state.json file existed. Verified via docker exec openclaw node -e 'const j=JSON.parse(require("fs").readFileSync("/home/node/.openclaw/cron/jobs.json","utf8")); console.log(Object.keys(j.jobs[0]).includes("state"))' → true. I did not determine the cause.
• Workaround attempted: none in place. The symlink workaround described in v2026.4.22: Telegram channels crash on startup — Cannot find package 'openclaw' in plugin-runtime-deps #70701 requires the plugin-runtime-deps/**/node_modules/ tree to exist; in our case the tree is not created because the npm install step fails first.
• Available on request: full /home/node/.npm/_logs/*-debug-0.log content, full docker logs openclaw from the upgrade window, openclaw.json excerpt, re-test against a fresh ~/.openclaw/ to determine whether pre-existing v2026.4.15 state is required for repro.