Bug type

Behavior bug (incorrect output/state without crash)

Summary

In a configured Telegram topic, OpenClaw showed typing and completed a final assistant message in the session transcript, but no reply was posted to Telegram.

Steps to reproduce

1. Start OpenClaw 2026.3.13 with Telegram forum topics configured.
2. In a configured Telegram topic, send a direct bot mention with a short question.
3. Observe Telegram showing the bot typing.
4. In the reproduced failed case, no final reply was posted to the topic even though the assistant session transcript contained a completed final answer.

Expected behavior

When a Telegram topic message is accepted and the assistant completes a final answer, the final reply should be posted to the same Telegram topic.

Actual behavior

Telegram showed the bot typing in topic 1312, the assistant session completed a final answer, and no reply was posted to Telegram.

Observed final generated answer in the session transcript was what I expected to be posted in the Telegram topic.

Observed session file:
/Users/userone/.openclaw/agents/main/sessions/aed48fc7-a4bd-4802-b64c-f2f3f665c8f1-topic-1312.json

Observed gateway error log near that time:
2026-03-19T16:52:29.536-07:00 [tools] tools.profile (coding) allowlist contains unknown entries (apply_patch).
No clear Telegram outbound send success/failure record was present before local instrumentation was added.

OpenClaw version

2026.3.13

Operating system

macOS 12.7.6

Install method

npm global

Model

google/gemini-2.5-flash

Provider / routing chain

openclaw -> google-generative-ai

Additional provider/model setup details

The observed failed session transcript recorded:

• provider: google
• modelApi: google-generative-ai
• model: gemini-2.5-flash

This was a Telegram forum topic conversation, topic id 1312, in a configured group chat.

Logs, screenshots, and evidence

Observed failed turn:

- User message time: `Thu 2026-03-19 16:52 PDT`
- Telegram topic: `1312`
- Session transcript file:
  - `/Users/userone/.openclaw/agents/main/sessions/aed48fc7-a4bd-4802-b64c-f2f3f665c8f1-topic-1312.jsonl`

The session transcript shows a completed final assistant answer:


My default GitHub scope is the `Interest-Driven-Life` organization: `https://github.com/orgs/Interest-Driven-Life/repositories`.

Gateway error log near that time: 
2026-03-19T16:52:29.536-07:00 [tools] tools.profile (coding) allowlist contains unknown entries (apply_patch). These entries are shipped core tools but unavailable in the current runtime/provider/model/config.

At the time of investigation, existing gateway logs did not clearly show:

outbound Telegram send attempted
outbound Telegram send succeeded
outbound Telegram send failed

Local follow-up instrumentation of the Telegram outbound send path made these stages visible.

Impact and severity

Affected users/systems/channels: Configured Telegram forum topics on this deployment
Severity: Medium
Frequency: Intermittent
Consequence: A user-visible bot reply can be silently lost even when the assistant completed generation

Additional information

This appears to be a post-generation outbound delivery observability problem in the Telegram send path, not a reasoning failure.

A local runtime patch that added Telegram outbound lifecycle logging for:

• typing start
• typing sent
• outbound start
• chunk/media send success
• final outbound success
• final outbound failure

made the delivery path diagnosable.

I currently have a local working patch that adds this instrumentation and can help turn it into a source-level PR if this direction is acceptable.