Require sandbox pass before builder deploy

a2a.yaml

@@ -1,5 +1,5 @@
 name: agent-builder
-version: 0.1.5
+version: 0.1.6
 entrypoint: agent:AgentBuilder
 description: Generate, test, and deploy new a2a agents from natural language.
 expose:

agent.py

@@ -49,7 +49,7 @@ class AgentBuilder(A2AAgent[BuilderConfig, NoAuth]):
         "Writes the project into the user's workspace, validates it in a "
         "sandbox, then ships it via the control plane."
     )
-    version = "0.1.5"
+    version = "0.1.6"
 
     config_model = BuilderConfig
     auth_model = NoAuth

agent_builder/tools.py

@@ -10,6 +10,7 @@ from __future__ import annotations
 import asyncio
 import ast
 import base64
+import hashlib
 import io
 import json
 import logging
@@ -465,6 +466,7 @@ def build_tools(ctx: ToolContext) -> list[Any]:
             return json.dumps({
                 "error": "no files in agents/" + name + "/ — write some first",
             })
+        source_hash = _source_bundle_hash(bundle_bytes)
         b64 = base64.b64encode(bundle_bytes).decode("ascii")
         script = (
             "set -e\n"
@@ -650,6 +652,7 @@ def build_tools(ctx: ToolContext) -> list[Any]:
             }
             if sanitized_fields:
                 result["sanitized_agent_card_fields"] = list(sanitized_fields)
+            _record_sandbox_result(store, name, source_hash=source_hash, result=result)
             return json.dumps(result)
 
         headers = _sandbox_headers(settings.sandbox_token, ctx.grant_token)
@@ -673,6 +676,7 @@ def build_tools(ctx: ToolContext) -> list[Any]:
         }
         if sanitized_fields:
             result["sanitized_agent_card_fields"] = list(sanitized_fields)
+        _record_sandbox_result(store, name, source_hash=source_hash, result=result)
         return json.dumps(result)
 
     @tool
@@ -716,11 +720,20 @@ def build_tools(ctx: ToolContext) -> list[Any]:
                 "error": "no CP JWT forwarded — agent declaration missing wants_cp_jwt=True",
             })
 
+        builder_state = _read_builder_state(store, name)
+        sanitized_fields = _sanitize_agent_workspace_pricing(store, prefix)
+        bundle_bytes = _tarball_workspace_dir(store, prefix)
+        if not bundle_bytes:
+            return json.dumps({"error": f"no files at agents/{name}/"})
+        source_hash = _source_bundle_hash(bundle_bytes)
+        sandbox_error = _sandbox_gate_error(builder_state, source_hash=source_hash)
+        if sandbox_error is not None:
+            return json.dumps(sandbox_error)
+
         try:
             latest_deploy = await _latest_cp_deployment(settings.cp_url, ctx.cp_jwt, name)
         except RuntimeError as exc:
             return json.dumps({"error": str(exc)})
-        builder_state = _read_builder_state(store, name)
         drift = _deployment_drift_error(
             name,
             latest_deploy,
@@ -733,10 +746,6 @@ def build_tools(ctx: ToolContext) -> list[Any]:
         if force:
             base_head = None
 
-        sanitized_fields = _sanitize_agent_workspace_pricing(store, prefix)
-        bundle_bytes = _tarball_workspace_dir(store, prefix)
-        if not bundle_bytes:
-            return json.dumps({"error": f"no files at agents/{name}/"})
         # Read the entrypoint from a2a.yaml so we can pass it verbatim.
         entrypoint = _read_entrypoint(bundle_bytes) or f"agent:{_class_name(name)}"
         try:
@@ -1268,6 +1277,55 @@ def _write_builder_state(
     store.put(_builder_state_key(name), body, "application/json")
 
 
+def _source_bundle_hash(bundle: bytes) -> str:
+    return hashlib.sha256(bundle).hexdigest()
+
+
+def _record_sandbox_result(
+    store: _ObjectStore,
+    name: str,
+    *,
+    source_hash: str,
+    result: dict[str, Any],
+) -> None:
+    state = _read_builder_state(store, name)
+    state["last_sandbox"] = {
+        "source_hash": source_hash,
+        "exit_code": result.get("exit_code"),
+        "updated_at": int(time.time()),
+    }
+    _write_builder_state(store, name, state)
+
+
+def _sandbox_gate_error(
+    builder_state: dict[str, Any],
+    *,
+    source_hash: str,
+) -> dict[str, Any] | None:
+    sandbox = builder_state.get("last_sandbox")
+    if not isinstance(sandbox, dict):
+        return {
+            "ok": False,
+            "error": "sandbox_not_passed",
+            "message": "Run test_agent_in_sandbox for the current source before deploying.",
+        }
+    if sandbox.get("source_hash") != source_hash:
+        return {
+            "ok": False,
+            "error": "sandbox_not_passed",
+            "message": "Source changed after the last sandbox pass; rerun test_agent_in_sandbox.",
+            "last_exit_code": sandbox.get("exit_code"),
+        }
+    if sandbox.get("exit_code") != 0:
+        return {
+            "ok": False,
+            "error": "sandbox_not_passed",
+            "message": "Last sandbox run failed; fix the source and rerun test_agent_in_sandbox.",
+            "last_exit_code": sandbox.get("exit_code"),
+        }
+    return None
+
+
 def _parse_manifest_json(raw: str) -> dict[str, Any]:
     try:
         parsed = json.loads(raw or "{}")

tests/test_template_init.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import asyncio
+import hashlib
 import io
 import json
 import tarfile
@@ -12,6 +13,7 @@ from agent_builder.builder import _workspace_backend_with_grep_compat
 from agent_builder.tools import (
     _BUILDER_STATE_FILE,
     _BUILDER_INTERNAL_PREFIX,
+    _WorkspaceObjectStore,
     _parse_manifest_json,
     _deploy_poll_url,
     _deployment_drift_error,
@@ -380,6 +382,18 @@ class TemplateInitTests(unittest.TestCase):
         }
         for rel, content in files.items():
             workspace.write_bytes(prefix + rel, content.encode("utf-8"))
+        source_hash = hashlib.sha256(
+            _tarball_workspace_dir(_WorkspaceObjectStore(workspace), prefix)
+        ).hexdigest()
+        workspace.write_bytes(
+            prefix + _BUILDER_STATE_FILE,
+            json.dumps({
+                "last_sandbox": {
+                    "source_hash": source_hash,
+                    "exit_code": 0,
+                }
+            }).encode("utf-8"),
+        )
         tools = build_tools(
             ToolContext(
                 bucket="bucket",
@@ -429,6 +443,41 @@ class TemplateInitTests(unittest.TestCase):
         self.assertEqual(dsl["entrypoint"]["module"], "agent")
         self.assertEqual(dsl["entrypoint"]["class_name"], "DemoAgent")
 
+    def test_cp_deploy_tarball_refuses_without_current_sandbox_pass(self) -> None:
+        workspace = _FakeWorkspace()
+        prefix = "agents/demo-agent/"
+        workspace.write_bytes(prefix + "agent.py", b"print('ok')\n")
+        workspace.write_bytes(
+            prefix + "a2a.yaml",
+            b"name: demo-agent\nversion: 0.1.0\nentrypoint: agent:DemoAgent\n",
+        )
+        workspace.write_bytes(prefix + "requirements.txt", b"")
+        tools = build_tools(
+            ToolContext(
+                bucket="bucket",
+                settings=_settings(deploy_wait_timeout_s=0),
+                workspace=workspace,
+                cp_jwt="jwt-user",
+            )
+        )
+        deploy = _tool_by_name(tools, "cp_deploy_tarball")
+
+        _FakeAsyncClient.posts = []
+        result = json.loads(
+            asyncio.run(
+                deploy.ainvoke(
+                    {
+                        "name": "demo-agent",
+                        "version": "0.1.0",
+                        "public": False,
+                    }
+                )
+            )
+        )
+
+        self.assertEqual(result["error"], "sandbox_not_passed")
+        self.assertEqual(_FakeAsyncClient.posts, [])
+
     def test_cp_compose_meta_agent_posts_manifest_and_records_state(self) -> None:
         workspace = _FakeWorkspace()
         tools = build_tools(