openwebui-regression-test
Skill0 starsChanged 56 days agoApache-2.0View on GitHub
Regression-test Open WebUI tools, adapters, streaming, reconnects, provider limits, and TTS through the real browser path.
In this repository12 other files it gives its agents.
- blackboardSkill
- chatterbox-tts-recoverySkill
- check-notes-firstSkill
- council-blackboardSkill
- invariant-guarded-debuggingSkill
- local-model-runtime-profilerSkill
- minimum-sufficient-changeSkill
- privileged-operationsSkill
- risk-aware-retrySkill
- session-handoverSkill
- symlink-space-saverSkill
- video-clip-editorSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
13 rules. About 0 are common, 13 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈13
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Use the real logged-in Open WebUI browser/app path for functional verification.
- Distinctive: line 37
Use timestamps or request IDs to join evidence across layers.
- Distinctive: line 41
Prefer the smallest fix at the layer that owns the fault.
- Distinctive: line 43
Run blocking synchronous tool work outside the event loop so Socket.IO heartbeats remain responsive.
- Distinctive: line 47
Keep intermediate tool-loop prose in structured receipts, but expose and speak only the final assistant answer.
- Distinctive: line 49
Avoid automatic Knowledge/RAG injection unless explicitly required.
- Distinctive: line 50
Do not modify account data or replace the live database to solve a request-path bug.
- Distinctive: line 54
After each material change:
- Distinctive: line 63
Check service restart counts and HTTP statuses.
- Distinctive: line 64
Remove only disposable test chats created by this test.
- Distinctive: line 65
Run database integrity checks and confirm the user's chat count/state remain plausible.
- Distinctive: line 69
Use the smallest matrix covering the changed behavior:
- Distinctive: line 95
Never claim “verified” from static configuration or a direct backend call when the bug occurs in the browser path.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Open WebUI Regression Test
Use the real logged-in Open WebUI browser/app path for functional verification. Treat configuration, service status, HTTP 200 responses, and database integrity as supporting evidence, not proof of user-visible success.
Establish the failure boundary
- Reproduce the user's exact model, prompt shape, enabled toggles, and tool path.
- Record whether the failure is:
- provider/API error such as HTTP 429;
- adapter or OpenClaw routing error;
- Open WebUI backend exception;
- Socket.IO heartbeat/reconnect warning;
- duplicated visible answer;
- missing or duplicated TTS;
- tool-call refusal, truncation, or repeated tool turns.
- Correlate browser-visible timing with service logs. Do not label a reconnect banner a backend crash unless a process restart or failed request proves it.
- Preserve the existing database and unrelated user state.
Inspect the complete request chain
Check, in order:
- Browser DOM and console-visible behavior.
- Open WebUI request and Socket.IO lifecycle.
- Open WebUI middleware/tool execution.
- Local adapter request/stream translation.
- OpenClaw gateway routing.
- Upstream provider response and rate-limit headers.
- Stored assistant message structure, including legacy
content, structuredoutput, tool receipts, and error fields. - TTS request and audio playback state when voice mode is involved.
Use timestamps or request IDs to join evidence across layers.
Apply narrow fixes
Prefer the smallest fix at the layer that owns the fault.
- Run blocking synchronous tool work outside the event loop so Socket.IO heartbeats remain responsive.
- For provider 429s, use bounded retries with server-directed or exponential backoff only when retrying is safe.
- Disable incidental title, tag, follow-up, or metadata model calls when they create avoidable concurrency or rate pressure.
- Batch independent retrieval calls in one parallel tool round when supported.
- Keep intermediate tool-loop prose in structured receipts, but expose and speak only the final assistant answer.
- Preserve tool calls, citations, and receipts; do not hide failures by discarding evidence.
- Avoid automatic Knowledge/RAG injection unless explicitly required.
- Do not modify account data or replace the live database to solve a request-path bug.
Validate through the real user path
After each material change:
- Restart only the services that require it.
- Open or reuse the actual logged-in Chrome/Open WebUI app profile.
- Submit a prompt that forces the affected behavior.
- Monitor the live DOM for reconnect banners, duplicated answer blocks, provider error cards, and completion.
- Confirm the expected tool calls occurred and inspect their stored receipts.
- Confirm the final stored message has no unexpected error and that visible
contentmatches the final answer once. - For voice mode, confirm a real TTS request returns valid audio and the app plays it unmuted.
- Check service restart counts and HTTP statuses.
- Remove only disposable test chats created by this test.
- Run database integrity checks and confirm the user's chat count/state remain plausible.
Test matrix
Use the smallest matrix covering the changed behavior:
- One ordinary response without tools.
- One forced single-tool response.
- One multi-tool or multi-search response.
- One streaming response from an OpenClaw-routed model.
- One provider likely to exercise the affected adapter path, such as NIM.
- Voice/TTS only when audio behavior changed.
For rate-limit fixes, test enough consecutive requests to exercise retry logic without deliberately hammering the provider.
Report receipts honestly
Report:
- exact browser path tested;
- model/provider;
- number and type of tool calls;
- reconnect banners observed;
- duplicate visible answers observed;
- upstream errors and retries;
- final request status;
- service restart evidence;
- database integrity and retained chat state;
- any untested boundary.
Never claim “verified” from static configuration or a direct backend call when the bug occurs in the browser path.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

