Agent Mesh
Go a2aa2a-protocolagent-meshagentic-aiinferencemcpmeshmesh-networknetwork-securitynetworking-aip2p-network
examples: substrate-actor: a second actor and a check for per-actor grants
Ignacio Janeiroc1697092acf6
diff --git a/development/examples/substrate-actor/Makefile b/development/examples/substrate-actor/Makefileindex 654cc594..a1bb6e4e 100644--- a/development/examples/substrate-actor/Makefile+++ b/development/examples/substrate-actor/Makefile@@ -7,6 +7,7 @@# make call NAME=demo from inside the actor: tools, calc allowed, payroll refused# make rogue an actor the mesh policy does not bind: refused# make policy push the mesh policy after editing manifests/mesh.yaml.tmpl+# make check two actors, different grants: assert each reaches only its own tool# make down remove the actors, the mesh and the provider# make clean delete the kind cluster#@@ -68,7 +69,7 @@ endifKUBECTL_CONTEXT := $(CTX)export KIND_CLUSTER_NAME KUBECTL_CONTEXT KO_DOCKER_REPO-.PHONY: cluster up issuer-docs actor call gemini rogue policy status down clean+.PHONY: cluster up issuer-docs actor call gemini rogue policy check status down cleancluster:@for t in docker kind ko kubectl go python3; do command -v $$t >/dev/null || { echo "$$t not found"; exit 1; }; done@@ -198,13 +199,29 @@ policy:--data-binary "$$($(KUBECTL) -n agentmesh get configmap mesh-config -o jsonpath='{.data.policy\.json}')"@echo+# check asserts the mesh keeps two authorized actors with different grants+# apart: demo (the `actor` role, mcp://calc) and ledger (the `ledger` role,+# mcp://payroll) each reach only their own tool. It pushes the policy first so+# the ledger binding is live, brings both actors up, then restarts the+# credential provider so its in-memory cache is empty and the actors' next+# request is minted under the policy just pushed, without waiting out+# --cache-ttl. The worker pool holds two actors, so a `rogue` left over from+# `make rogue` must be suspended first.+check:+ $(MAKE) policy+ $(MAKE) actor NAME=demo+ $(MAKE) actor NAME=ledger+ $(KUBECTL) -n agentmesh rollout restart deployment/credential-provider+ $(KUBECTL) -n agentmesh rollout status deployment/credential-provider --timeout=90s+ ROUTER_PORT=$(ROUTER_PORT) ATESPACE=$(ATESPACE) KUBE_CONTEXT=$(CTX) ./check.sh+status:$(KUBECTL) -n agentmesh get pods$(KATE) get actors -a $(ATESPACE)$(KUBECTL) -n ate-system logs deployment/atenet-egress -c ext-proc --tail=20 2>/dev/null | grep -iE 'agentmesh|credential|denied' || truedown:- -for a in demo rogue; do $(KATE) suspend actor $$a -a $(ATESPACE) >/dev/null 2>&1; $(KATE) delete actor $$a -a $(ATESPACE) >/dev/null 2>&1; done+ -for a in demo ledger rogue; do $(KATE) suspend actor $$a -a $(ATESPACE) >/dev/null 2>&1; $(KATE) delete actor $$a -a $(ATESPACE) >/dev/null 2>&1; done# The namespace and the credential-provider Service stay: the egress gateway# was installed with that Service's address and holds its ClusterIP.-$(KUBECTL) -n agentmesh delete deployment,configmap,secret,networkpolicy --all --timeout=120sdiff --git a/development/examples/substrate-actor/README.md b/development/examples/substrate-actor/README.mdindex 3c870702..357f9193 100644--- a/development/examples/substrate-actor/README.md+++ b/development/examples/substrate-actor/README.md@@ -24,10 +24,20 @@ make call NAME=demo # tools, calc allowed, payroll refumake rogue # an actor the policy does not bind: refusedmake gemini NAME=demo GCP_PROJECT=<project> # Gemini on Vertex AI through the mesh; see the guidemake policy # push the policy after editing manifests/mesh.yaml.tmpl+make check # demo and ledger, different grants: each reaches only its own toolmake down # the actors and the agentmesh namespacemake clean # the kind cluster```+`make check` runs two actors bound to different roles: `demo` to `actor`+(`mcp://calc`) and `ledger` to `ledger` (`mcp://payroll`). `check.sh` prints+one line per case and exits non-zero on a mismatch. A refusal counts only+when the serving node refuses the call (`mcp.py` exit 3), not when the mesh+offers the actor no provider at all (exit 4). `make check` restarts the+credential provider so the policy it pushes applies at once instead of+waiting out the cache. The worker pool of the demo template holds two actors,+so suspend `rogue` before `make check`.+The provider has its own Go module (`credential-provider/go.mod`) thatdepends on the Substrate release it builds against. `go test ./...` in thatdirectory runs its unit tests without a cluster.diff --git a/development/examples/substrate-actor/check.sh b/development/examples/substrate-actor/check.shnew file mode 100755index 00000000..b5671ad4--- /dev/null+++ b/development/examples/substrate-actor/check.sh@@ -0,0 +1,98 @@+#!/usr/bin/env bash+# Copyright 2026 Google LLC+#+# Licensed under the Apache License, Version 2.0 (the "License");+# you may not use this file except in compliance with the License.+# You may obtain a copy of the License at+#+# http://www.apache.org/licenses/LICENSE-2.0+#+# Unless required by applicable law or agreed to in writing, software+# distributed under the License is distributed on an "AS IS" BASIS,+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.+# See the License for the specific language governing permissions and+# limitations under the License.++# Two authorized actors with different grants, and the mesh keeps them apart.+# demo is bound to the `actor` role (mcp://calc); ledger to the `ledger` role+# (mcp://payroll). The check asserts that each reaches only its own tool and+# that the other tool is refused by the serving node, not skipped by the+# client.+#+# `make check` is the turnkey entry point: it pushes the policy, brings both+# actors up and restarts the credential provider so its cache is empty before+# calling this script. To run it on its own, first make sure the+# ledger binding is live and both actors are running and holding a credential+# minted under it:+#+# make up+# make actor NAME=demo+# make actor NAME=ledger # the worker pool holds two actors; suspend a+# # third (make rogue leaves `rogue` suspended)+# make policy # push the ledger binding to the control plane+# kubectl -n agentmesh rollout restart deployment/credential-provider+# # drop the provider cache so the next mint uses+# # the policy just pushed (or wait out --cache-ttl)+# ./check.sh # or: make check, which does all of the above+#+# mcp.py exits 0 when the tool answers, 3 when a provider is reached and the+# node refuses the call, and 4 when the mesh offers this actor no provider at+# all. A real isolation deny is 3: the node saw the call and said no.+set -uo pipefail+cd "$(dirname "${BASH_SOURCE[0]}")" || exit 1++ROUTER_PORT="${ROUTER_PORT:-8000}"+ATESPACE="${ATESPACE:-ate-demo-sandbox}"+KUBE_CONTEXT="${KUBE_CONTEXT:-kind-kind}"+KUBECTL=(kubectl --context "${KUBE_CONTEXT}")++if (exec 3<>"/dev/tcp/127.0.0.1/${ROUTER_PORT}") 2>/dev/null; then+ echo "127.0.0.1:${ROUTER_PORT} is in use; free it or set ROUTER_PORT" >&2; exit 1+fi+"${KUBECTL[@]}" -n ate-system port-forward service/atenet-router "${ROUTER_PORT}:80" >/dev/null 2>&1 &+pf=$!; trap 'kill $pf 2>/dev/null' EXIT INT TERM HUP+# The port-forward is backgrounded and silenced, so a failed start (wrong+# context, namespace, or no permission) would otherwise surface only as+# confusing call errors below. Poll the port until it accepts a connection,+# bounded (5s) and aborting early if the process dies, instead of a fixed sleep.+ready=false+for _ in {1..50}; do+ if (exec 3<>"/dev/tcp/127.0.0.1/${ROUTER_PORT}") 2>/dev/null; then ready=true; break; fi+ kill -0 "$pf" 2>/dev/null || break+ sleep 0.1+done+if [[ "${ready}" != "true" ]]; then+ echo "kubectl port-forward to atenet-router did not start or bind the port" >&2+ exit 1+fi++failures=0+# expect NAME WANT-EXIT ACTOR SERVICE: run one call and check its exit code.+# 0 allowed, 3 refused by the serving node, 4 no provider offered.+expect() {+ local name=$1 want=$2 actor=$3 service=$4 out got+ out=$(ROUTER_PORT="${ROUTER_PORT}" ATESPACE="${ATESPACE}" python3 mcp.py add "${actor}" "${service}" 2 3 2>&1)+ got=$?+ local tail="${out##*$'\n'}"+ if [[ "${got}" == "${want}" ]]; then+ printf ' PASS %-40s exit %s %s\n' "${name}" "${got}" "${tail}"+ else+ printf ' FAIL %-40s got exit %s, want %s %s\n' "${name}" "${got}" "${want}" "${tail}"+ failures=$((failures + 1))+ fi+}++echo "== each actor reaches its own tool"+expect "demo -> calc (its grant)" 0 demo calc+expect "ledger -> payroll (its grant)" 0 ledger payroll+echo "== neither can borrow the other's grant (the serving node refuses)"+expect "demo -> payroll (not its grant)" 3 demo payroll+expect "ledger -> calc (not its grant)" 3 ledger calc++echo+if [[ "${failures}" -eq 0 ]]; then+ echo "OK: the two actors are isolated."+else+ echo "FAIL: ${failures} case(s) did not match."+fi+exit $(( failures > 0 ? 1 : 0 ))diff --git a/development/examples/substrate-actor/manifests/mesh.yaml.tmpl b/development/examples/substrate-actor/manifests/mesh.yaml.tmplindex c87d564b..a571e2fe 100644--- a/development/examples/substrate-actor/manifests/mesh.yaml.tmpl+++ b/development/examples/substrate-actor/manifests/mesh.yaml.tmpl@@ -19,10 +19,14 @@ data:{"name": "mesh:role:router", "allowed_services": ["*"], "allowed_targets": ["*"]},{"name": "actor","allowed_services": ["system://mesh.catalog", "mcp://calc"],+ "allowed_targets": ["*"]},+ {"name": "ledger",+ "allowed_services": ["system://mesh.catalog", "mcp://payroll"],"allowed_targets": ["*"]}],"bindings": [- {"role": "actor", "members": ["user:spiffe://substrate-actor.local/actor/ate-demo-sandbox/demo"]}+ {"role": "actor", "members": ["user:spiffe://substrate-actor.local/actor/ate-demo-sandbox/demo"]},+ {"role": "ledger", "members": ["user:spiffe://substrate-actor.local/actor/ate-demo-sandbox/ledger"]}]}office.yaml: |diff --git a/development/examples/substrate-actor/mcp.py b/development/examples/substrate-actor/mcp.pyindex 3c099834..cc2ed7fe 100755--- a/development/examples/substrate-actor/mcp.py+++ b/development/examples/substrate-actor/mcp.py@@ -93,14 +93,20 @@ def add(actor, service, a, b):print(f" (no tool of mcp://{service} is visible to this actor; calling the office node directly)")peers = {r["peer_id"] for r in rpc(actor, "discover_remote_services", {"type": "mcp"})[0] or []}if not peers:+ # The mesh shows this actor no provider of the service: no role binds+ # its identity to it, so discovery returns nothing and no node ever+ # sees a call. Exit 4 keeps this apart from a server-side refusal (3),+ # so a caller cannot mistake "never asked" for "asked and denied".print(f"mcp://{service}/add({a}, {b}) -> refused: the mesh shows this actor no MCP provider at all (no role binds its identity)")- sys.exit(3)+ sys.exit(4)peer = sorted(peers)[0]result, text = rpc(actor, "call_remote_tool", {"peer_id": peer, "tool_name": f"mcp://{service}/add","arguments": {"a": float(a), "b": float(b)},})if result is None:+ # A provider was reached and the serving node refused the call. This is+ # the enforced deny: exit 3, distinct from the no-provider case above.print(f"mcp://{service}/add({a}, {b}) -> refused: {text}")sys.exit(3)print(f"mcp://{service}/add({a}, {b}) = {text}")