openclaw-lighthouse

Telegram no-reply after a specific prompt (exo man build mac 3 ultra cluster)

Summary

A user reported that sending a prompt like:

give me the original link about that exo man build mac 3 ultra cluster

may cause OpenClaw to stop replying in Telegram.

Current evidence is a symptom report (not yet a confirmed root cause). This note tracks the pattern so future incidents can be compared quickly.

Environment

• First report date: 2026-02-24
• Channel: Telegram (direct chat)
• OpenClaw version: not captured in the report
• Model/provider at incident time: not captured

Reproduction

1. In Telegram, send the prompt:
   • give me the original link about that exo man build mac 3 ultra cluster
2. Observe whether the bot replies.
3. If no reply, send a simple follow-up prompt (for example: hi).
4. Check whether reply flow resumes or remains silent.

Expected vs actual

• Expected: bot replies normally to the prompt and follow-up messages.
• Actual: in some reports, no reply is sent after this prompt; follow-up may also stay silent.

Findings

1. This is currently a reported symptom, not a confirmed deterministic trigger.
2. Outward behavior (“Telegram stops replying”) overlaps with other reliability issues, including model-chain failures and channel/session state drift.
3. Without incident-time logs, we cannot separate:
   • model timeout/cooldown exhaustion,
   • channel delivery path issues,
   • transient session/runtime stalls.

Mitigation / Workaround

When this happens, capture evidence immediately:

openclaw logs --follow
openclaw status --all
openclaw gateway status
openclaw channels status --probe

Operational workarounds:

1. Retry with a short split query (break one long request into 2 messages).
2. Send a simple health-check prompt (hi) to test reply path.
3. If still silent, restart gateway and retest:
   • openclaw gateway restart
4. If recurrence is frequent, keep a timestamped incident log and compare against provider cooldown/auth errors.

Risk / Impact

• User impact: appears as random “bot went silent” behavior.
• Ops impact: difficult to triage after the fact when logs are not captured live.
• Trust impact: reliability confidence drops, even if root cause is transient.

Related Issues/PRs

• OpenAI Codex account switch (Plus → Pro) + onboard reset: Telegram stops replying
• All models failed: fallback chain exhausted by cooldown/auth failures
• OpenAI Codex provider cooldown: all models failed (no available auth profile)

Next actions

• Reproduce with live logs on the same host/channel path.
• Check whether this prompt is a true trigger or just correlated with provider/channel instability.
• If reproducible, open upstream issue with redacted logs + exact timestamp.

References

• Troubleshooting
• Gateway CLI
• Channels
• Model Failover