training loop

[Bug]: Why tools.profile (coding) allowlist contains unknown entries ?

2026.3.14, tools.profile (coding) allowlist contains unknown entries (apply_patch, image). These entries are shipped core tools but unavailable in the current runtime/provider/model/config. What matter? Why?

[Bug]: openclaw --profile <name> devices list fails with gateway closed (1000) / handshake timeout on 2026.3.13

When using a non-default profile / secondary gateway, `openclaw --profile <name> devices list` fails with: gateway connect failed: Error: gateway closed (1000): [openclaw] Failed to start CLI: Error: gateway closed (1000 normal closure): no close reason

[Skills] 37 "Skipping skill path that resolves outside its configured root" warnings from built-in extensions

```markdown

Ollama fails with TLS error

[Bug]: v0.17.0 4*2080ti 22G Qwen3.5 RPC call to sample_tokens timed out.

Responses API → Chat Completions bridge does not translate `developer` role to `system` for non-OpenAI providers

Ollama desktop app respawns ollama.exe serve after manual process termination on Windows

[Bug/Troubleshoot] Permission Denied / File does not exist

localePrefix: "as-needed" causes 308 redirects on locale subpaths in Next.js 16 (Turbopack) - middleware never called

[Bug]: SessionManager.fileEntries grows unbounded in memory, causing 1.4GB+ heap growth in long-running sessions

There's a slow leak hiding in the persistence layer that most people are working around without knowing what's causing it. `SessionManager.fileEntries` is an in-memory array that mirrors every line written to the JSONL session transcript. Every message, tool call, tool result, and metadata entry gets pushed to it. Nothing ever removes them. Over hours of normal use, this array quietly accumulates hundreds of thousands of objects, eventually holding 1.4GB+ of retained heap. V8 can't collect any of it because SessionManager holds a live reference to the array for the lifetime of the process. The tricky part: the LLM context window stays bounded. `replaceMessages()` keeps `Agent._state.messages` at a manageable size during compaction. So from the model's perspective, everything looks fine. But underneath, `fileEntries` is a parallel data structure in the persistence layer that compaction never touches. Two layers, only one with a ceiling. This is likely the root cause behind multiple reported memory and session-bloat issues, including #6190, #13758, #2254, and #20910, which all share the same symptom pattern. It's a contributing factor in several others (see Related Issues below).

[Bug]: `vertex_ai/gemini-embedding-2-preview` routes to `:predict` endpoint, returns `FAILED_PRECONDITION`

[Bug]: Gemma 4 E4B extremely slow on v0.19.0 forced TRITON_ATTN fallback yields ~9 tok/s on RTX 4090 (vs ~100+ tok/s for comparable Llama 3B)

[Feature]: Multi-provider credential pools for cross-provider failover and rotation

[Bug]: vllm start Qwen3.5-35B-A3B fail

[Bug]: Docker: Matrix plugin fails to load with "Cannot find module '@vector -im/matrix-bot-sdk'" (Require stack: /app/dist/matrix-*.js) — persistent even after cleanup/rebuild

The Matrix plugin fails to load in Docker (`openclaw:local` or built images) with the persistent error: [matrix] [default] channel exited: Cannot find module '@vector -im/matrix-bot-sdk' Require stack:/app/dist/matrix-ZSlmDGn7.js Even after: - Deleting `/app/extensions/matrix`, `/app/dist/matrix-*.js`, and `/home/node/.openclaw/extensions/matrix` - Running `openclaw plugins install @openclaw/matrix` - Running `npm install --omit=dev` in both locations - Multiple `--no-cache` rebuilds using a custom Dockerfile with `OPENCLAW_EXTENSIONS=matrix` - Cleaning the config with `openclaw doctor --fix` and manual resets The loader keeps hitting the incomplete bundled version shipped in the base image. The user-installed copy never fully takes precedence, and the runtime is missing `@vector-im/matrix-bot-sdk` (and sometimes other shared modules). ai helped me write this report after a day of trying to get matrix plugin installed and working properly in openclaw. would be nice to use a real messenging protocol and element instead of the channels of telegram and signal.

[Bug]: session_send does not work for sandboxed agents

After enabling sandbox mode for all my agents except the main agent, my agents are no longer able to send messages to each other. `sessions_send` does not work (times out). The non-sandbox agent (main) does not have this issue. This suggests a sandbox-specific regression in session cross-session routing.

[Bug]: Too many arguments error when running openclaw browser extension install/path

root@YANHUA:~/.openclaw# openclaw browser extension install Config was last written by a newer OpenClaw (2026.3.31); current version is 2026.3.24. error: too many arguments for 'browser'. Expected 0 arguments but got 2. root@YANHUA:~/.openclaw# openclaw browser extension path Config was last written by a newer OpenClaw (2026.3.31); current version is 2026.3.24. error: too many arguments for 'browser'. Expected 0 arguments but got 2. root@YANHUA:~/.openclaw# I can control browser via Managed Browser or Remote CDP, but when installing the browser extension, it thows out the error message above.

[Bug] Compaction fails intermittently with extra usage requirement error

message tool: 'No active WhatsApp Web listener' error despite channel showing connected

The \message\ tool consistently returns \Error: No active WhatsApp Web listener (account: default)\ when attempting to send outbound messages, even though inbound messages are flowing normally and \openclaw channels status\ reports the WhatsApp channel as connected.

v2026.3.31: Dashboard returns 500 Internal Server Error on GET /

After updating to 2026.3.31, the dashboard at http://127.0.0.1:18789/ returns 500 Internal Server Error on every request. The gateway starts correctly (Telegram bots connect, WebSocket works, TUI works), but the HTTP handler crashes before serving any response.