What installing gives you
The install is one name over five distributions and twenty-three import packages, of which a reader touches about five and six do not import at all on a bare setup.
Last updated: 14 Oct, 2026 · Agent Governance Toolkit 4.1.0
The setup lesson installed the toolkit with one command. That command pulled in far more than the course uses. Knowing what landed, and what quietly failed to import, saves an hour of confusion later.
Listing the core packages
The twenty-three core import packages carry these names. Most import; a handful raise, because they need an optional dependency a bare install does not pull in, or because of a bug in the package itself.
core = ["agent_compliance","agent_control_plane","agent_os","agent_os_observability",
"agent_primitives","agent_runtime","agentmesh","amb_core","atr","caas",
"cmvk","emk","hypervisor","iatp","mcp_kernel_server","nexus",
"agent_sandbox","agent_sre","mcp_trust_server",
"a2a_agentmesh","agent_mcp_governance","mcp_receipt_governed","mcp_trust_proxy"]Counting what imported
Trying to import each name and sorting the results shows the real picture at a glance.
import importlib
# import each package, split the names into those that load and those that raise
ok, fail = [], []
for name in core:
try:
importlib.import_module(name)
ok.append(name)
except Exception as e:
fail.append((name, str(e).split("No module named")[-1].strip()))
print(f"packages {len(core)} import OK {len(ok)} import FAIL {len(fail)}")
for name, why in fail:
print(f" {name:22} missing {why}")packages 23 import OK 17 import FAIL 6 agent_os_observability missing 'prometheus_client' amb_core missing 'aiofiles' cmvk missing 'numpy' emk missing 'numpy' iatp missing 'fastapi' agent_mcp_governance missing 'agent_os.governance'
What the failures mean
- Five failures are missing optional dependencies.
prometheus_client,aiofiles,numpyandfastapiare not pulled in by a bare install; the course never needs those five packages. - One failure is a real bug.
agent_mcp_governancetries to importagent_os.governance, a module that does not exist in 4.1.0. No dependency fixes it. - Everything the course uses is in the seventeen that import. You do not need the broken ones.
The five packages the course touches
| Package | What you use it for | Taught in |
|---|---|---|
agent_os | govern(), YAML policy, the JSONL audit logger | govern(): the smallest real gate and Writing a policy in a YAML file |
agent_control_plane | intercept_tool_execution, the policy engine, the flight recorder | One choke point for every tool call |
agentmesh | agent identity, delegation, the Merkle audit chain | An agent that can prove who it is |
mcp_trust_proxy | governing an MCP server | A trust proxy for an MCP server |
agent_sre | circuit breakers and an error budget | Circuit breakers and an error budget |
Where the import map helps
- Debugging an ImportError on a fresh machine, to tell a missing optional dependency from a real bug.
- Trimming a deployment, installing only the layers you use rather than every package.
- Reading the toolkit's own code, knowing which packages are live and which are stubs.
ImportError from one of these packages does not mean your install is broken. Five of them need a dependency the course does not use, and one is broken in the release itself. Import what you need and leave the rest.Related
- Install
numpyand re-run the check. Watchcmvkandemkmove into the OK list. - Import
agent_mcp_governanceon its own and read the full traceback that names the missing module. - Print the OK list and find the five packages the table says the course uses.
Little by little, you're building something great.