nexus-llm-adapters
Skill158 starsChanged 11 days agoMITView on GitHub
How to add or debug a Nexus LLM provider adapter — what every adapter must wire, how to verify it against a live provider, and the failure modes that produce a silently blank chat. Use when adding or changing a provider adapter, debugging streaming or reasoning display, working with local or CLI-backed models, or touching chat branch plumbing.
In this repository13 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- nexus-agentsSkill
- nexus-eval-harnessSkill
- nexus-mobile-compatSkill
- nexus-model-evalSkill
- nexus-model-updatesSkill
- nexus-releaseSkill
- nexus-storageSkill
- nexus-testingSkill
- nexus-tool-schemasSkill
- nexus-ui-mockupsSkill
- nexusSkill
The file
Files
Nexus LLM adapters
Adapters live in src/services/llm/adapters/{provider}/ and extend BaseAdapter,
which talks direct HTTP through ProviderHttpClient. There are no vendor LLM SDKs
in this repo; adding one is a decision to raise, not to make.
This router points at the procedure. The detail loads when you take the path.
Workflow
- Adding or changing an adapter — follow
protocols/add-adapter.md. It is a wiring checklist; work it in order rather than pattern-matching an existing adapter, because the pieces that get forgotten are the ones no compiler checks. - Chat is blank, missing reasoning, looping, or doubling — follow
protocols/debug-adapter.md. Start from the symptom, not the adapter source: these failures have known causes and reading code first wastes the trip. - Before calling any adapter change done — run
protocols/verify-adapter.md. You MUST exercise the non-streaming path, the streaming path, and the streaming error path separately; they fail independently and unit tests prove none of them. You MUST also runscripts/check_stream_error_wiring.pyon the adapter you touched and get a zero exit. - End of a session that used this skill — run
protocols/self-refine.md.
Map
protocols/the procedures: add-adapter, debug-adapter, verify-adapter, self-refine.references/mechanism and rationale, read on demand:symptoms.md(the symptom→cause lookup the debug protocol enters through),streaming-contract.md,reasoning-rendering.md,local-providers.md,cli-providers.md,chat-plumbing.md.scripts/check_stream_error_wiring.py— checks that an adapter parsing an SSE stream wiresextractError. Run it, do not re-derive it by reading source.
Siblings — do not duplicate them here
nexus-model-updatesowns model metadata: ids, pricing, context windows, capability flags, provider defaults, and the live provider smoke test.nexus-model-evalandnexus-eval-harnessown grading models on tool use.nexus-testingowns test lanes, what a mock can prove, and the in-app loop.nexus-mobile-compatowns Node imports, dependency vetting, and desktop-only gating — which every adapter that shells out or touches the filesystem needs.
Refine
At the end of a session that used this skill, run protocols/self-refine.md and
append to refinement-log.md.
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.

