Add Grok model context limits (131K tokens)

- Added 'grok' pattern to MODEL_SPECIFIC_LIMITS with 131_072 token limit
- This fixes the issue where CLI was showing 0/32000 tokens for Grok models
- All Grok models (grok-3, grok-3-mini, etc) now correctly show 0/131072 tokens
- Added debug logging to xAI provider for better troubleshooting

Author: jack

.goose/memory/project improvement ideas.txt

@@ -24,3 +24,24 @@ Successfully added xAI provider support for Grok models to Goose:
 **Branch:** feature/add-grok-models
 **Commit:** ff54e6d576
 
+# ["xai", "fix", "endpoint", "url", "api"]
+Fixed critical issue with xAI provider endpoint URL:
+
+**Problem:** 
+Getting 404 error when using xAI provider because of incorrect API endpoint construction.
+
+**Root Cause:**
+The code was constructing URL as: `https://api.x.ai` + `v1/chat/completions` = `https://api.x.ai/v1/chat/completions`
+But x.ai API documentation shows the base URL should be `https://api.x.ai/v1` and then append `chat/completions`.
+
+**Solution:**
+1. Changed `XAI_API_HOST` from `"https://api.x.ai"` to `"https://api.x.ai/v1"`
+2. Updated URL joining to append `"chat/completions"` instead of `"v1/chat/completions"`
+3. Updated UI provider registry default host to match
+4. Tested with curl - now returns proper 400 auth error instead of 404
+
+**Additional Issue Found:**
+User had `GOOSE_LEAD_MODEL: claude-opus-4-20250514` in config which was triggering lead/worker mode. This model doesn't exist, causing cascade failures. Recommend removing or updating to valid model.
+
+**Status:** Fixed and committed (commit 30dbcba3e4)
+

crates/goose/src/model.rs

@@ -30,6 +30,9 @@ static MODEL_SPECIFIC_LIMITS: Lazy<HashMap<&'static str, usize>> = Lazy::new(||
     // Meta Llama models, https://github.com/meta-llama/llama-models/tree/main?tab=readme-ov-file#llama-models-1
     map.insert("llama3.2", 128_000);
     map.insert("llama3.3", 128_000);
+
+    // x.ai Grok models, https://docs.x.ai/docs/overview
+    map.insert("grok", 131_072);
     map
 });
 

crates/goose/src/providers/xai.rs

@@ -79,6 +79,9 @@ impl XaiProvider {
             ProviderError::RequestFailed(format!("Failed to construct endpoint URL: {e}"))
         })?;
 
+        tracing::debug!("xAI API URL: {}", url);
+        tracing::debug!("xAI request model: {:?}", self.model.model_name);
+
         let response = self
             .client
             .post(url)