logging issue

`openclaw update` fails with "Completion cache update failed (qa scenario pack not found)"

When running `openclaw update`, the **Completion cache update** step produces the following error: ``` Completion cache update failed ([openclaw] Failed to start CLI: Error: qa scenario pack not found: qa/scenarios/index.md at readQaScenarioPack (file:///usr/local/lib/node_modules/openclaw/dist/suite-BW4kSK9C.js:754:27) at readQaScenarioById (file:///usr/local/lib/node_modules/openclaw/dist/suite-BW4kSK9C.js:790:19) at readQaScenarioExecutionConfig (file:///usr/local/lib/node_modules/openclaw/dist/suite-BW4kSK9C.js:795:9) at readRequiredDiscoveryRefs (file:///usr/local/lib/node_modules/openclaw/dist/suite-BW4kSK9C.js:2834:9) at file:///usr/local/lib/node_modules/openclaw/dist/suite-BW4kSK9C.js:2840:39 ```

[Regression 0.20.x] memory layout cannot be allocated for all models >10 GB on Windows — GPU abandoned after vision encoder CPU buffer allocation failure

[BUG] Constantly hitting the "conversation is too long" error message with new chats despite only 16k/200k (8%) tokens used

[Bug]: openclaw onboard --mode local --no-install-daemon fails with shouldNormalizeGoogleProviderConfig is not a function, while node dist/index.js onboard ... succeeds

openclaw onboard --mode local --no-install-daemon fails before entering onboarding with loadFacadeModule(...).shouldNormalizeGoogleProviderConfig is not a function, while node dist/index.js onboard --mode local --no-install-daemon enters the wizard successfully under the same installation.

devices approve deadlock after gateway restart — node-host repair request cannot be approved via CLI

After a gateway restart following the 2026.4.1 update, the local node-host service requests a role-upgrade repair pairing (roleFrom=operator roleTo=node). This creates an unresolvable deadlock: openclaw devices list works (uses local fallback) openclaw devices approve <requestId> returns unknown requestId — no local fallback path openclaw doctor --fix does not detect or resolve the pending request The Control UI dashboard shows no pending pairing request to approve via UI All CLI commands requiring gateway auth fail with 1008: pairing required The result is a fully locked gateway that cannot be recovered through any supported CLI path.

[Bug] Claude Code incorrectly billing against extra usage instead of plan quota

When flash attention is not supported, quantized KV cache should be disregarded instead of aborting the model run.

[Bug] v2026.3.11: Discord thread context re-injected every turn (context pollution + cache bust)

Ollama 0.20.2 unknown model architecture: 'gemma4' with Cuda arch 50

[Bug]: NVIDIA Provider Sends Anthropic-Style Message Format to OpenAI-Compatible API

When spawning a subagent with `sessions_spawn` using an NVIDIA provider configured as `"api": "openai-completions"`, OpenClaw sends Anthropic-style message content (arrays with `type` objects) instead of OpenAI-style simple strings. This causes the NVIDIA API to reject requests with HTTP 400 errors. bug was created using openclaw *

Gateway service: exit code mismatch causes StartLimitBurst=5 exhaustion on rapid restart

trainer.evaluate() fails after trainer.train()

[Bug]: tools[].custom.eager_input_streaming: Extra inputs are not permitted — Bedrock rejects custom tool fields for Haiku 4.5

native ollama-go-engine: TurboQuant implementation

[Bug]: WS node pairing auto-approved at runtime but never persists (device-auth.json never created)

OpenClaw 2026.3.8 (3caab92) Setup Mac node host via SSH tunnel to local gateway Worker: 127.0.0.1:18791 → Gateway: 127.0.0.1:18789 Confirmed - tunnel works (nc -vz 127.0.0.1 18791 succeeds) - worker node service runs - worker has ~/.openclaw/identity/device.json - worker never creates ~/.openclaw/identity/device-auth.json Gateway behavior gateway logs show: device pairing auto-approved device=df72580fdf4e6ddf4a8635f50ecc2c23870a247ab1e0608feef328e4be1be34c role=node Node runtime state: Known: 1 · Paired: 0 · Connected: 1 Research Node ... unpaired · connected Device store: openclaw devices list --json { "pending": [], "paired": [ { "nodeId": "cb91b74b0afad0605fb8725ec991d86cccd24b3e059a73966a902778e19a8156" } ] } So the node reaches the runtime registry but never enters the persistent device pairing store and never writes device-auth.json. Question Why does WS node pairing auto-approve at runtime but never persist to devices pending/paired or create device-auth.json on the worker? 'Known: 1 · Paired: 0 · Connected: 1 Research Node ... unpaired · connected'

bug: multiple issues in anthropic /v1/messages experimental pass-through to openai

There are five related bugs in the Anthropic `/v1/messages` experimental pass-through implementation when routing requests to OpenAI/Azure models. These issues were discovered during integration testing with Claude Code CLI which uses `input_text` content block format and requires forcing routing through Chat Completions API when the downstream proxy doesn't support Responses API.

i cant use ollama on rx6600m

[Bug]: Qwen3.5-27B fails to start with CPU KV cache offloading (`--kv_offloading_backend native`) while Qwen3-32B works fine

Bug: Gateway ignores baseUrl for ollama2, routes all requests to ollama port

When configuring multiple Ollama providers (`ollama` on port 11434 and `ollama2` on port 11435), the Gateway routes ALL requests to the first provider (`ollama` on port 11434), completely ignoring the `baseUrl` configured for `ollama2`.

ROCm backend fails to initialize on AMD Radeon AI PRO R9700 (RDNA4, gfx1201) in Windows 11