Goal

Configure Telegram multi-account routing so a specific account (for example luna) is bound to an independent agent, without breaking the main bot.

Hard Rules

1. Never claim success based only on config write or restart.
2. Always distinguish:

• config write success
• service restart success
• routing/deployment success

3. Validate against current installed version behavior, not assumed docs.
4. On OpenClaw 2026.3.2, do not place routing under agents.bindings.
5. Preserve main bot availability.
6. Prefer supported CLI workflows over speculative manual JSON edits.

Trigger Conditions

• The user wants to add a child/sub agent to Telegram with its own bot account.
• A Telegram account is healthy in status/health but still routes to agent:main:*.
• The user wants one OpenClaw gateway to host multiple Telegram bots mapped to different agents.
• A previous attempt used the wrong config layer and broke routing.

Execution Procedure

1. Baseline checks:

openclaw config validate
openclaw status --deep
openclaw health --json

2. If config is invalid due to misplaced routing config such as agents.bindings, remove the invalid key and revalidate.
3. Backup config file:

cp ~/.openclaw/openclaw.json ~/.openclaw/openclaw.json.bak_$(date +%Y%m%d-%H%M%S)

4. Confirm the target agent exists:

openclaw agents list --json

5. If needed, create or prepare the target Telegram account under channels.telegram.accounts.<accountId>.
6. Create routing via the supported CLI:

openclaw agents bind --agent <agentId> --bind telegram:<accountId> --json

7. Verify the binding persisted:

openclaw agents bindings --json
openclaw agents list --bindings --json

8. Restart gateway:

openclaw gateway restart

9. Re-check service health:

openclaw status
openclaw health --json

10. Verify outbound delivery from the target Telegram account:

openclaw message send --channel telegram --account <accountId> --target <chatId> --message "<probe>" --json

11. Verify inbound routing proof:

• Ask the user to reply to the probe text from Telegram.
• Inspect sessions/logs and prove the session key lands under agent:<agentId>: rather than agent:main:.

Evidence Checklist

The final report must include all of the following:

1. Binding object exists (channel=telegram, accountId=<id>, agentId=<agent>).
2. Config validates.
3. Gateway restarted successfully.
4. Outbound message from the target account succeeded.
5. Inbound message routed to the target agent session key.
6. Main bot still healthy.

Failure Modes and Handling

• If agents.bindings appears on OpenClaw 2026.3.2: remove it and switch to the supported CLI routing workflow.
• If docs conflict with runtime: trust installed runtime + CLI behavior.
• If route proof is unavailable because the user has not replied yet: mark the setup as pending external inbound verification; do not claim fully complete.

Output Contract

Always output:

1. Root cause or setup objective.
2. Exact commands executed.
3. What changed in config or routing state.
4. Verification evidence with concrete outputs.
5. Remaining blocker (if any).