From 83253916c514062375790158cd07a8ef6ccbe19e Mon Sep 17 00:00:00 2001
From: Kevin Meisel <kevin.meisel03@gmail.com>
Date: Tue, 3 Feb 2026 21:39:52 +0100
Subject: [PATCH] Fix eval prompt collision with model flag

---
 src/upskill/cli.py | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)

diff --git a/src/upskill/cli.py b/src/upskill/cli.py
index 6e8c680..82df596 100644
--- a/src/upskill/cli.py
+++ b/src/upskill/cli.py
@@ -45,12 +45,31 @@
 
 console = Console()
 
+KNOWN_PROVIDERS = ("anthropic", "openai", "generic")
+
+
+def _model_has_provider(model: str) -> bool:
+    return any(model.startswith(f"{provider}.") for provider in KNOWN_PROVIDERS)
+
+
+def _apply_provider_to_models(models: list[str], provider: str | None) -> list[str]:
+    if not provider:
+        return models
+    normalized: list[str] = []
+    for model in models:
+        if _model_has_provider(model):
+            normalized.append(model)
+        else:
+            normalized.append(f"{provider}.{model}")
+    return normalized
+
 
 @asynccontextmanager
 async def _fast_agent_context() -> AsyncIterator[object]:
     fast = FastAgent(
         "upskill",
         ignore_unknown_args=True,
+        parse_cli_args=False,  # Prevent FastAgent from consuming upskill CLI flags (e.g. -m).
     )
 
     @fast.agent()
@@ -719,6 +738,8 @@ async def _eval_async(
     if not models:
         models = [config.effective_eval_model]
 
+    models = _apply_provider_to_models(models, provider)
+
     is_benchmark_mode = len(models) > 1 or num_runs > 1
 
     async with _fast_agent_context() as agent: