#18867: fix: route slug generator LLM call through configured provider

## Problem `generateSlugViaLLM()` calls `runEmbeddedPiAgent()` without passing `provider` or `model`, so the embedded agent defaults to a direct Anthropic API call. For users routing through LiteLLM (or other proxies), this silently fails — no API key for Anthropic — and every session falls back to a timestamp slug (e.g. `0231`) instead of a descriptive one (e.g. `slug-bug`). ## Fix Extract the provider and model from `agents.defaults.model.primary` (e.g. `litellm/claude-opus-4-6` → provider `litellm`, model `claude-opus-4-6`) and pass them to `runEmbeddedPiAgent()`. This ensures the slug generator uses the same model routing as the rest of the app. ## Testing Before fix: every `/new` produced timestamp slugs (`2026-02-17-0210.md`, `2026-02-17-0220.md`, etc.) After fix: `/new` produces descriptive slugs (`2026-02-17-slug-bug.md`) <!-- greptile_comment --> <h3>Greptile Summary</h3> Fixes slug generation for users routing LLM calls through LiteLLM or other proxies by extracting provider and model from `agents.defaults.model.primary` and passing them to `runEmbeddedPiAgent()`. Without this fix, the slug generator defaults to direct Anthropic API calls, which fail for proxy users without Anthropic keys, resulting in timestamp-based slugs instead of descriptive ones. <h3>Confidence Score: 4/5</h3> - This PR is safe to merge with minimal risk - The fix correctly addresses the root cause by extracting provider/model from config and passing to the embedded agent. The parsing logic is sound (using indexOf + slice), though it differs from the existing splitModelRef utility in subagent-spawn.ts. The fallback behavior (undefined when no slash found) matches the function signature expectations. - No files require special attention _{Last reviewed commit: 15e3c16} <!-- greptile_other_comments_section --> _{(2/5) Greptile learns from your feedback when you react with thumbs up/down!} <!-- /greptile_comment -->