fix: use platform llm grants for builder

README.md

@@ -47,7 +47,7 @@ Three platform-managed bits land on the inner skill via `RunContext`:
 | Field | Source | Required? |
 |---|---|---|
 | `ctx.workspace.bucket` | grant minted by the orchestrator | yes |
-| `ctx.llm` | caller's LLM creds (Card declares `llm_provisioning=caller_provided`) | yes — the LLM bill goes to the caller |
+| `ctx.llm` | platform-scoped LiteLLM grant (Card declares `llm_provisioning=platform`) | yes |
 | `ctx.cp_jwt` | caller's CP JWT (Card declares `wants_cp_jwt=True`) | yes — used by `cp_deploy_tarball` |
 
 The user opts into all three when they pick `agent-builder` from the
@@ -55,8 +55,8 @@ marketplace — the dashboard already surfaces these on the agent card.
 
 ## Pricing
 
-`$0.10 / call`, plus your LLM cost via your own provider. The deployed
-agent is yours forever; subsequent invocations of YOUR agent run on
+`$0.10 / call` during the current pricing model. The deployed
+agent is yours forever; subsequent invocations of your agent run on
 their own pricing (whatever you set when generating it).
 
 ## How to call it

agent.py

@@ -10,8 +10,8 @@ This agent only works when invoked through the platform orchestrator,
 because it needs three things forwarded from the caller:
 
   1. ``ctx.workspace.bucket`` — the user's MinIO bucket
-  2. ``ctx.llm`` — caller's LLM creds (this is an expensive skill;
-     the user pays for inference directly via their own provider)
+  2. ``ctx.llm`` — a platform-scoped LiteLLM grant token for the builder's
+     own DeepAgents loop
   3. ``ctx.cp_jwt`` — the user's CP JWT, so cp_deploy_tarball can
      POST to /v1/agents/from-tarball as the user.
 """
@@ -53,15 +53,18 @@ class AgentBuilder(A2AAgent[BuilderConfig, NoAuth]):
     config_model = BuilderConfig
     auth_model = NoAuth
     tools_used = ("deepagents", "langgraph", "a2a-pack", "litellm", "microsandbox")
-    llm_provisioning = LLMProvisioning.CALLER_PROVIDED
+    llm_provisioning = LLMProvisioning.PLATFORM
     wants_cp_jwt = True
     workspace_access = WorkspaceAccess.dynamic(
         allowed_modes=(WorkspaceMode.READ_ONLY, WorkspaceMode.READ_WRITE_OVERLAY),
     )
     pricing = Pricing(
         price_per_call_usd=0.10,
-        caller_pays_llm=True,
-        notes="LLM cost passes through; deployed agent is yours forever.",
+        caller_pays_llm=False,
+        notes=(
+            "Uses a platform-scoped LiteLLM grant for the builder's own "
+            "reasoning loop; deployed agent is yours forever."
+        ),
     )
 
     @skill(

agent_builder/builder.py

@@ -27,8 +27,9 @@ class BuilderContext:
     project_prefix: str | None = None
     workspace: Any | None = None
     grant_token: str | None = None
-    # Optional caller-provided LLM creds (when the outer platform Card
-    # declares llm_provisioning=caller_provided). Falls back to settings.
+    # LLM creds forwarded by the orchestrator according to this agent's Card.
+    # The builder itself uses platform grants; generated agents usually use
+    # caller-provided creds. Falls back to settings for local development.
     llm_base_url: str | None = None
     llm_api_key: str | None = None
     llm_model: str | None = None

tests/test_builder_prompt.py

@@ -24,6 +24,13 @@ class BuilderPromptTests(unittest.TestCase):
         )
         self.assertIn("stream=True", agent_source)
 
+    def test_outer_builder_uses_platform_llm_grants(self) -> None:
+        agent_source = Path(__file__).resolve().parents[1].joinpath("agent.py").read_text()
+
+        self.assertIn("llm_provisioning = LLMProvisioning.PLATFORM", agent_source)
+        self.assertIn("caller_pays_llm=False", agent_source)
+        self.assertIn("platform-scoped LiteLLM grant", agent_source)
+
     def test_builder_defaults_raise_timeout_budget(self) -> None:
         config_source = Path(__file__).resolve().parents[1].joinpath(
             "agent_builder/config.py"