Bug: openclaw status / openclaw status --json hang or die under 2026.4.21 due to heavy status runtime/plugin load path

Summary

On OpenClaw 2026.4.21 (f788c88), openclaw status and openclaw status --json are unreliable in this environment.

Symptoms:

• command hangs for a long time
• child openclaw process grows very large in memory
• command path often ends in SIGKILL

By contrast, these are healthy:

• openclaw --version
• openclaw gateway status --json
• gateway service itself
• gateway RPC/health paths

This suggests the bug is in the broader status runtime/plugin loading path rather than in the gateway probe path.

Version

• OpenClaw: 2026.4.21 (f788c88)

Environment

• OS: Fedora Linux
• Install root: /home/asistente/.nvm/versions/node/v24.14.1/lib/node_modules/openclaw
• OpenClaw config path: ~/.openclaw/openclaw.json
• Gateway managed via user systemd service

Reproduction

Run:

openclaw --version
openclaw gateway status --json
openclaw status --json

Expected behavior

openclaw status --json should return structured JSON in a reasonable amount of time.

Actual behavior

• openclaw gateway status --json succeeds
• openclaw status --json stalls for a long time
• during the run, a child openclaw process grows large in RSS
• the failing runs often terminate with SIGKILL

What was ruled out

1. Session store size / sessions.json

Not the primary cause in this environment.

Observed:

• ~/.openclaw/agents/main/sessions/sessions.json was only ~267 KB
• only ~10 session entries

2. Gateway probe path

Not the issue.

Observed:

• openclaw gateway status --json consistently returns valid structured output
• gateway health and RPC remain good

3. Model config/auth drift

There was a separate config drift after upgrade:

• openai/gpt-5.4 was configured as primary
• runtime auth was actually openai-codex/gpt-5.4

This was corrected, but openclaw status --json remained defective afterwards.

Investigation details

The problem appears specific to the heavier status path.

From code inspection, status --json goes through the broader status scan and plugin/runtime loading path. During testing, memory growth was real and large enough to look like a runtime loading problem rather than a simple network timeout.

Observed examples during failing runs:

• child process RSS in the hundreds of MB
• gateway/service memory peaks around ~1G to ~1.3G during some restart/test cycles

Plugin isolation tests

To test whether a single heavy plugin was responsible, I temporarily disabled likely heavy contributors and re-ran the check.

Test A: disable browser

• restart gateway
• run openclaw status --json

Result:

• memory usage improved somewhat
• command still hung / remained defective

Test B: disable browser + acpx

• restart gateway
• run openclaw status --json

Result:

• runtime plugin set shrank further
• command still hung / remained defective

Interpretation:

• browser and acpx do contribute runtime weight
• but neither appears to be the sole root cause
• issue likely sits in the general plugin/runtime loading path used by status

Config was restored afterwards.

Likely root cause

A regression in the general plugin/runtime loading path used by:

• openclaw status
• openclaw status --json

rather than a simple failure in:

• gateway probing
• session summary alone
• a single plugin only

Current workaround

Use:

openclaw --version
openclaw gateway status --json

And use in-product session/runtime status surfaces instead of relying on openclaw status in this version/environment.

Additional note

A same-version refresh (openclaw update) reported:

• Before: 2026.4.21
• After: 2026.4.21

So this issue persisted even after a package-manager refresh + doctor + daemon restart of the same release.