docs(auto_routing): update the setup skill and page for tier labels, classifier prompt, and fallback

[devin-ai-integration]

Summary

Four complexity-router changes landed in litellm after the skill was written (PR #772 merged 2026-08-05), so the skill was telling agents about a config surface that no longer matches the code:

• tier_labels renames the four tiers for the dashboard, spend logs, and the LLM classifier rubric while config keys stay canonical (#35893)
• classifier_llm_config.system_prompt replaces the built-in rubric outright, with GET /auto_router/classifier/default_prompt to fetch the real default text (#35855)
• classifier_fallback: heuristic | default_model decides what runs when the classifier call fails (#35855)
• reminder_markers overrides the marker pair whose blocks get stripped before classification (#35874)

static/skills/auto-router keeps its shape: still an interview that asks for the router name and each tier model, still presents the proxy's own models as a numbered menu, still closes by laying out the defaults in force. New tier-names and reminder-block rows join that defaults table, the LLM-classifier follow-up now asks about the failure path and offers (never assumes) a replacement rubric, and step 6 gained a pre-flight against POST /auto_router/test_routing, which classifies a prompt for a config that only exists in a file without creating the router or calling the routed model.

docs/proxy/auto_routing.md changed only where the code contradicted it. "Falls back to the heuristic scorer if the LLM classifier errors" is now conditional on classifier_fallback; the section claiming to show "every knob v2 exposes" gained the four new keys, with tier_labels commented out so the canonical tier names stay the visible baseline and renaming reads as opt-in; the reminder-stripping sentence notes the pair is overridable; and the UI section described a form that no longer exists, since Add Auto Router now opens on name plus Template with everything else under a collapsed Detailed Configuration, and carries Test Routing next to Test Connection (#35746, #35859).

Deliberately left alone: the short "To get started, tell your agent" block stays the only place the setup flow appears on the page; litellm_settings.autorouter_savings_baseline_model and the cost-optimization savings work (#35521, #35907) are spend reporting rather than router setup; and the CJK keyword-match fix (#35984) plus the embedding context-window fix (#35956) change behavior without changing any documented key.

npm run build passes; no new warnings.

Link to Devin session: https://app.devin.ai/sessions/081e8ba97c494972aebf20b0d63eda61

---

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview	Aug 5, 2026 11:19pm

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 1 additional finding.