my openclaw says this after /new
One of "input" or "previous_response_id"or 'prompt'or 'conversation_id' must be provided.
#help
grand crescent
wise lavaBOT
vale thunderBOT
That error is coming from the model/provider request, not from /new itself. /new should reset the session cleanly, so this smells like either a version bug or the Discord native slash-command path sending an empty request.
First check, please paste:
openclaw --version
openclaw status --all
Also try /reset as a standalone message. If Discord is opening the native slash-command UI, try sending it as plain text instead of selecting the app command.
Docs: https://docs.openclaw.ai/concepts/session and https://docs.openclaw.ai/channels/discord#native-commands-and-command-auth
grand crescent
telegram and discord
vale thunderBOT
Got it — if it happens in both Telegram and Discord, it’s probably not Discord-specific. It sounds like /new is reaching the agent/provider with an empty request instead of being handled as a reset.
Please paste:
openclaw --version
openclaw status --all
Then reproduce once and check:
openclaw logs --follow
Also try /reset as a standalone message. If both /new and /reset fail on the latest version, this is likely a bug/regression worth filing: https://github.com/openclaw/openclaw/issues
Ask Kaan to run:
openclaw status --all
That includes the OS, OpenClaw version, gateway status, and a redacted log tail — much more useful than just the OS here.
grand crescent
Diagnosis (read-only)
Gateway connection details:
Gateway target: ws://127.0.0.1:18727
Source: local loopback
Config: /home/kaanbot/.openclaw/openclaw.json
Bind: loopback
✓ Config: /home/kaanbot/.openclaw/openclaw.json
✓ Secret diagnostics (0)
✓ Restart sentinel: none
✓ Port 18727
Detected dual-stack loopback listeners (127.0.0.1 + ::1) for one gateway process.
! Tailscale exposure: serve · daemon unknown · vmd191245.tail928d63.ts.net
https: https://vmd191245.tail928d63.ts.net
✓ Skills: 14 eligible · 0 missing · /home/kaanbot/.openclaw/workspace
✓ Plugin compatibility (none)
✓ Channel issues (none)
Pasteable debug report. Auth tokens redacted.
Troubleshooting: https://docs.openclaw.ai/troubleshooting
◇
vale thunderBOT
This report is useful: the gateway is reachable and there are no channel issues, so this probably isn’t Telegram/Discord routing.
It’s also probably Linux/VPS based on /home/kaanbot, but the report doesn’t show the exact distro/version.
Next useful bits:
openclaw --version
openclaw models status
Then reproduce /new once and immediately run:
openclaw logs --limit 300 --plain
Paste the part around that exact error. The error text looks like an OpenAI/Responses provider request with an empty payload, so the logs + version should tell us whether this is a known regression, bad model config, or stale session/runtime state.
The Tailscale warning is worth reviewing later, but it’s probably not causing this /new error. Docs: https://docs.openclaw.ai/cli/logs
grand crescent
2026-04-29T18:11:17.452Z error [tools] read failed: Path escapes sandbox root (~/.openclaw/workspace): /home/kaanbot/self-improving/heartbeat-state.md raw_params={"path":"/home/kaanbot/self-improving/heartbeat-state.md","limit":200}
2026-04-29T18:11:17.454Z error [tools] read failed: Path escapes sandbox root (~/.openclaw/workspace): /home/kaanbot/proactivity/heartbeat.md raw_params={"path":"/home/kaanbot/proactivity/heartbeat.md","limit":200}
2026-04-29T18:11:17.455Z error [tools] read failed: Path escapes sandbox root (~/.openclaw/workspace):
2026-04-29T18:11:17.456Z error [tools] read failed: Path escapes sandbox root (~/.openclaw/workspace): /home/kaanbot/proactivity/session-state.md raw_params={"path":"/home/kaanbot/proactivity/session-state.md","limit":200}
2026-04-29T18:11:17.458Z error [tools] read failed: Path escapes sandbox root (~/.openclaw/workspace): /home/kaanbot/proactivity/session-state.md raw_params={"path":"/home/kaanbot/proactivity/session-state.md","limit":200}
2026-04-29T18:35:50.421Z info gateway/ws {"subsystem":"gateway/ws"} ⇄ res ✓ channels.status 912ms conn=112f3789…4fe3 id=50fc483c…690e
2026-04-29T18:35:52.250Z info gateway/ws {"subsystem":"gateway/ws"} ⇄ res ✓ health 445ms conn=80d92e92…3804 id=0b9307e7…89c5
["error_handling","lifecycle","agent_end","assistant_error"],"runId":"8baa1b1f-94d4-4c2b-ab77-25a71af86ac5","isError":true,"error":"One of "input" or "previous_response_id"or 'prompt'or 'conversation_id' must be provided.","failoverReason":null,"model":"gpt-5.5","provider":"openai-codex","rawErrorPreview":"One of "input" or "previous_response_id"or 'prompt'or 'conversation_id' must be provided.","rawErrorHash":"sha256:f4df43227553","providerRuntimeFailureKind":"unclassified"} embedded run agent end
2026-04-29T18:35:58.802Z info gateway/channels/telegram {"subsystem":"gateway/channels/telegram"} telegram sendMessage ok chat=6696108665 message=2035
vale thunderBOT
The important line is this one:
provider="openai-codex" model="gpt-5.5"
error="One of \"input\" or \"previous_response_id\" ... must be provided."
So yeah: Telegram/Discord are fine. This is failing inside the OpenAI Codex provider run.
The earlier Path escapes sandbox root lines are separate: the agent is trying to read /home/kaanbot/... files outside its workspace (~/.openclaw/workspace). Worth fixing later, but not the /new provider error.
Next paste these please:
openclaw --version
openclaw models status