Apigee Agent Gateway Governance
GoogleOfficialPublished by Google
- Skill
- Workflow
- Automation and integrations
- Apache-2.0
- Low risk
- Has scripts
AutoGineer summary
Written by AutoGineer from the skill’s own files, not by the author
This skill automates the configuration of an Apigee API proxy to act as a service extension for a Google Cloud Agent Gateway. It enables inspection of LLM traffic to perform token metering, caller identity extraction, prompt screening via Model Armor, and semantic caching. The skill follows a structured deployment process using AFT templates and gcloud commands, ensuring the proxy is correctly wired to intercept and evaluate agent requests without acting as a backend router.
What it does
- Stands up an inspect-only Apigee proxy for LLM traffic governance.
- Configures token metering per model and enforces usage quotas.
- Integrates Model Armor for prompt screening and semantic caching.
- Manages gateway network connectivity and AuthzPolicy configuration.
- Provides automated deployment scripts for consistent environment setup.
- Supports both development and production modes with defined contract schemas.
When to use it
- When you need to meter or govern LLM token usage for agents.
- When you require prompt screening or semantic caching for agent egress.
- When you need to extract caller identity from agent traffic.
- When you are wiring Apigee as an AI gateway behind a Google Cloud Agent Gateway.
Advantages
- Centralized governance for agent model egress.
- Version-controlled proxy configuration via AFT templates.
- Automated handling of complex network and service extension dependencies.
- Non-intrusive inspection architecture that does not require backend routing.
Usage examples
- “Set up Apigee governance for my existing Agent Gateway in dev mode.”
- The agent will assess the current environment, apply default configurations for metering and identity, and generate a deployment plan for approval.
- “Deploy the Apigee proxy using my production configuration file.”
- The agent will validate the provided contract file, execute the deployment sequence, and verify the proxy is reporting as EXTENSIBLE.
Install
Download the skill as a ZIP:
https://autogineer.com/agent-skills/google/apigee-agent-gateway-governance/downloadUnzip it into the cross-client skills folder of your project:
mkdir -p .agents/skills && unzip apigee-agent-gateway-governance.zip -d .agents/skills
Documentation by Google
Apigee as an Agent Gateway Service Extension
Wires an Apigee proxy in as an ext-proc callout on a GCP Agent Gateway. The proxy is a processor, not an API proxy: it has no backend, inspects the LLM request and response the gateway hands it, and returns a verdict.
Usage Attribution
MUST READ: Before executing any gcloud, REST API, or MCP command, you
MUST read and follow the instructions in
Usage Attribution.
You build the proxy; the user does not.
Two modes, one schema. Decide which before anything else, and say which.
Input file present → prod mode: it is the contract, every value comes from
it, and a missing or malformed field is a refusal, not a default. No input file
and the user is describing an outcome → dev mode: apply the documented
defaults, show a plan with durations and get approval, start with identity and
metering only, write apigee-agent-gateway-governance.yaml at the end as the
handoff. assets/apigee-agent-gateway-governance.example.yaml documents the
schema both modes use.
Modes, defaults, durations, cost disclosure and the interlocks that bind both:
references/modes.md. Read it before you start.
Colour the plan's ACTION column — green REUSE, cyan CREATE/RESERVE,
yellow APPEND/MODIFY; full palette in references/modes.md. Yellow is the
one that earns its place: it is the only signal that a resource the user already
owns is being changed. Rendered plain, the one row that modifies something is
indistinguishable from the dozen that do not — which is exactly the row the
approval exists for.
Prerequisite: where the agent runs
Check this before building anything. It decides whether traffic can reach the gateway at all, and it is silent when wrong — every step below still succeeds while nothing arrives.
Use Agent Runtime, created with agent_gateway_config /
agent_to_anywhere_config. It is the supported path and the orchestrator builds
the client-side PSC, routing, DNS and cert trust for you. There is no gcloud
surface for this and no console path — it is SDK-only, at agent creation time.
See
Route traffic through Agent Gateway
for the binding, and references/networking-psc.md for a worked example.
Set GOOGLE_API_PREVENT_AGENT_TOKEN_SHARING_FOR_GCP_SERVICES: "False" in the
agent's env_vars, or every invocation 401s on the agent's own
sessions.create, before user code runs. Deploy-time invisible. Full symptom
and isolation: references/gotchas.md, Agent Gateway.
Where the agent runs is otherwise settled — the premise is a working gateway and
a working agent. Do not explain how to build one or list unsupported compute
unprompted; references/networking-psc.md has the constraints if it turns out
to matter.
Tool versions
Do not assume these are installed or current.
| Tool | Minimum | Why |
|---|---|---|
gcloud | 588.0.0 | apigee apis import --from-template is on the GA track at this version; older SDKs need gcloud beta |
google-cloud-aiplatform[agent_engines,adk] | >=2.1,<3 | Agent Runtime client used to bind the agent to the gateway |
google-adk | >=2.9,<3 | only if you build the test agent with ADK |
On gcloud >= 588.0.0: every feature must keep its uid, and the template must
never declare endpoints or targets. All six features set uid today and
the template declares neither, so the shipped bundle is correct — but both are
load-bearing now, and breaking either one still compiles, still deploys, and
reports nothing. Mechanism and symptoms: references/gotchas.md, under
Authoring (AFT).
gcloud version | head -1 # confirm >= 588.0.0
gcloud components update # if older
pip install "google-cloud-aiplatform[agent_engines,adk]>=2.1,<3" "google-adk>=2.9,<3"scripts/compile_check.py needs PyYAML and names the interpreter if it is
missing. pip install pyyaml, or run it with a Python 3 that has it; gcloud's
bundled interpreter does not, and the SDK's vendored yaml is Python 2.
python3 scripts/compile_check.py assets/apigee-proxy-template.yamlExecution model
Three layers, and the choice between them is not stylistic:
| Layer | Used for | Why |
|---|---|---|
| AFT YAML template | the proxy itself | version-controlled, reviewable, diffable |
gcloud | PSC, DNS, AuthzExtension, AuthzPolicy | native commands exist |
REST (curl) | env properties, envgroups, deployments, debug sessions | no gcloud equivalent today |
The customer owns one input file
(assets/apigee-agent-gateway-governance.example.yaml). It is the
version-controlled artifact. Never hand-edit the generated proxy: regenerate it.
Run the Apigee-side sequence with the script, not by hand. Steps 1, 3 and 4
below are order-dependent and each fails in a way that reads as a different
problem. scripts/deploy_proxy.sh does all three, preserves the properties
map, polls the async operation, and exits non-zero if the deployment does not
come up EXTENSIBLE:
scripts/deploy_proxy.sh --org "$ORG" --env "$ENV" --proxy "$PROXY" \
--template assets/apigee-proxy-template.yaml [--service-account "$SA"] --yes \
[--index-endpoint-id ID --index-id ID --sc-project P --sc-region R --embeddings-model M]Run it without --yes first: it prints the plan and changes nothing. Show that
plan, then re-run with --yes once the user agrees.
The vector-search flags are the one exception to "edit the default in every
feature that declares it". Those three ids are substituted at deploy time
and checked for leftovers; hand-editing them into feature-semantic-cache.yaml
deploys fine, skips the check, and makes the file single-use.
scripts/ensure_gateway_connectivity.sh is the same idea for the gateway leg
(step 0). It is the only part of this skill that modifies a resource the user
already owns, so it applies nothing without --yes.
The steps are written out below so you can diagnose a failure, not so you can retype them.
Inputs you must resolve first
Environment prerequisites: references/setup.md — gcloud SDK (588+),
project, billing, gcloud services enable, region, credentials, IAM. Nothing
below creates any of it.
Prod: from the input file, then environment, then ask — never invent. Dev: apply
the references/modes.md defaults, and ask only for what has none.
project,regionapigee.org,apigee.environment,apigee.envgroup,apigee.hostname,apigee.proxynetwork.vpc,network.subnet,network.networkAttachmentgateway.name,gateway.matchPaths(a list, one entry per wire shape),gateway.manageNetworkConfig
matchPaths drives TWO artifacts; write both from the same list. The
AuthzPolicy paths[].suffix gets one suffix per entry, verbatim with
leading punctuation kept — it decides what the gateway routes to Apigee. The
template's LLM_URI_MATCH gets one (request.uri Like "*<entry>*") clause
per entry, OR-ed, leading : or / stripped — it decides what the proxy
acts on once it arrives. The shipped two-entry default gives:
((request.uri Like "*generateContent*") or (request.uri Like "*chat/completions*"))The AuthzPolicy is the outer gate and wins: a suffix list narrower than
matchPaths means the missing shape never reaches Apigee at all. Derive both
from the input file and never from the default when the file disagrees — the
file is the reviewable contract and a mismatch is invisible once deployed.
A one-entry list is legitimate and governs one shape. Dropping an entry
governs one fewer silently — deploys clean, zero transactions for the shape
dropped — and narrows feature-normalize, Model Armor and the cache together.
references/gotchas.md, Runtime.
semanticCache is a KNOWN ISSUE whenever it is on — warn, then build it.
The key is the prompt embedding alone: no caller, no model, no shape. A hit
therefore crosses callers — similar prompts share an answer — and one model
behind one shape does not make it safe. With more shapes the first model to
answer answers for the other too: empty HTTP 200 across shapes, wrong-model
body within one. Warn whenever the cache is on, not only on multi-shape, then
build. Wording: references/modes.md; mechanism: references/gotchas.md.
Each entry is a SUFFIX match. You write these, not the user — derive them
from which endpoint their agent actually calls, and do not ask them to reason
about suffixes. Vertex's OpenAI-compatible endpoint and api.openai.com need
different entries, and getting it wrong governs nothing silently; the forms and
the evidence are in references/gotchas.md, Runtime.
There is no apiShape field. Metering does not depend on the wire shape —
the count is normalised in JS across both — so the field selected nothing and
was removed. Adding a shape is an || branch in extract-model.js plus an
entry in matchPaths.
Three places carry the callout hostname and all three must match — the
Apigee envgroup, the AuthzExtension authority/service, and the private DNS A
record. You set all three from one value: a build invariant to hold, not a
question for the user. It becomes a topic only when a callout never lands, and
then it is the first check. Build order matters alongside it — the
AuthzExtension is inert until the host resolves and the proxy is deployed, so
"the extension exists" is not evidence that leg works. references/gotchas.md,
Runtime.
Workflow
0. Gateway: give it a route into the VPC
A gateway with no networkConfig has no route into the customer's VPC and
cannot resolve the callout hostname. This skill builds it — network
attachment, egress and DNS peering — so its absence is a build step, not a
blocker, and should be reported that way. Do not open with an alarm about a
resource you are about to create.
One thing here does need the user, and it is the reason this is step 0 rather than a silent prerequisite: attaching egress is one-way and affects traffic they already run. Settle it in the plan, before anything executes — see "Confirm everything up front" below. Capturing a known-good agent turn first is worth doing when they are nervous about a production gateway, because it separates "never worked" from "we broke it"; it is a precaution to offer, not a step to impose.
gcloud network-services agent-gateways describe "$GATEWAY" \
--location="$REGION" --project="$PROJECT" --format='value(networkConfig)'
gcloud network-security authz-policies list \
--location="$REGION" --project="$PROJECT"A console-created gateway already has a CUSTOM AuthzPolicy — an IAP
REQUEST_AUTHZ one, which coexists with ours and is not needed for the egress
JWT. You never create the gateway, so assess it rather than assuming.
references/gotchas.md, Agent Gateway.
Empty networkConfig output means missing, and it is easy to miss: the gateway
still serves ordinary agent traffic perfectly, because that path goes to public
Google endpoints and never enters the VPC. Only the Apigee callout fails.
If it is missing, behaviour depends on gateway.manageNetworkConfig:
true— run the script below. It creates the network attachment, then read-modify-writes the gateway to add the egress attachment and DNS peering. Agreement for this is taken in the plan, not here;--yescarries it.false— the user has said they want to make this change themselves. Print the script invocation and let them run it. Do not modify the gateway.
scripts/ensure_gateway_connectivity.sh \
--gateway "$GATEWAY" --project "$PROJECT" --region "$REGION" \
--vpc "$VPC" --subnet "$SUBNET" --attachment "$ATTACHMENT" \
--domain "$DNS_DOMAIN" # add --yes only after the user agreesEither way networkConfig has to exist before the Apigee work starts — not
because its absence is an error, but because every later step then succeeds and
governs nothing.
Use the script, not the raw commands: import is a whole-resource upsert and a
hand-built YAML silently drops the gateway's registries and protocols. Why,
and the etag and PROTOCOL_UNSPECIFIED handling: references/gotchas.md.
1. Apigee: enable the environment
The env must carry apigee-service-extension-enabled=true. No gcloud flag
exists; PATCH the properties map.
curl -s -H "Authorization: Bearer $TOKEN" \
"https://apigee.googleapis.com/v1/organizations/$ORG/environments/$ENV"The PATCH replaces the whole map. Read it first and re-send every property you intend to keep, or you will silently drop them.
curl -X PATCH -H "Authorization: Bearer $TOKEN" -H "Content-Type: application/json" \
"https://apigee.googleapis.com/v1/organizations/$ORG/environments/$ENV?updateMask=properties" \
-d '{"properties":{"property":[{"name":"apigee-service-extension-enabled","value":"true"}]}}'This does not disable Apigee Analytics — the two coexist; if analytics looks
missing it is the separate per-environment add-on. references/gotchas.md.
Use a dedicated environment, attached to exactly one envgroup. An
environment in two makes the gateway 404 every request with the extension never
invoked. It also leaves the / base path free, which is what the proxy ships
with.
scripts/ensure_apigee_environment.sh creates the environment, attaches it to
the regional runtime instance, and creates and attaches the envgroup — so step 2
is already done when it succeeds. It creates billable resources and applies
nothing without --yes; that agreement is taken in the plan, not here.
scripts/ensure_apigee_environment.sh \
--org "$ORG" --env "$ENV" --region "$REGION" \
--envgroup "$ENVGROUP" --hostname "$HOSTNAME" # add --yes to applyIt is idempotent and resumable, and refuses with exit 2 when the environment is
already in another envgroup or the envgroup exists on a different hostname. It
also handles three things a runbook tends to get wrong — the omitted type on
create, the four-minute instance attachment that locks the environment, and
polling the operation rather than a GET that succeeds too early.
2. Apigee: envgroup on the callout hostname
Done already if step 1's script ran. Verify, or do it by hand for an envgroup that already exists. The hostname must be exactly the callout host, and the group must be attached to the env.
curl -s -H "Authorization: Bearer $TOKEN" \
"https://apigee.googleapis.com/v1/organizations/$ORG/envgroups/$ENVGROUP/attachments"How many envgroups is the environment in? More than one and the callout
never lands. deploy_proxy.sh enforces this before touching anything and exits
2 naming them; assess_integration.sh reports it. Read the failure literally if
it fires — the environment is wrong, not the script.
3. Build the proxy from the template
gcloud apigee apis import organizations/$ORG/apis/$PROXY \
--from-template=assets/apigee-proxy-template.yamlEach features: entry resolves relative to the template file's own
directory, never the working directory, so a bare filename must sit beside the
template (a subpath like features/feature-identity.yaml works too). assets/
is flat by convention, not compiler demand — but moving a file without updating
features: fails the import with a bare FileNotFoundError.
If you get Permission denied on resource ... (or it may not exist) on an org
you demonstrably have access to, it is the credential, not the access. With
context-aware access enabled the default gcloud credential is rejected. Pass the
ADC token instead:
CLOUDSDK_AUTH_ACCESS_TOKEN=$(gcloud auth application-default print-access-token) \
gcloud apigee apis import ...To change a parameter, edit the default in every FEATURE that declares it
— not in the template. gcloud apigee apis import passes an empty
caller-parameter map, so each feature resolves its own default and the
template's declaration of the same name is never read. Editing the template
changes nothing, silently, and the compile still passes. Measured: setting
MA_CONTINUE_ON_ERROR to DRY_RUN in the template shipped ENFORCE on a
fail-closed policy. LLM_URI_MATCH is declared in four features and all four
must agree. scripts/compile_check.py fails on any disagreement. There is no
--parameters flag.
Enable an optional recipe by adding its feature to the template's features
list. See references/recipes.md.
tokenMetering.models in the input file is NOT generative. It documents
which buckets exist; it does not create them. A bucket is real only when it has
an LLMTokenQuota step in feature-token-metering.yaml and a matching
*_TOKEN_LIMIT parameter. Three exist: gemini-pro, gemini-flash and other
— the catch-all for any model whose name contains neither pro nor flash.
Adding a fourth entry to the list meters nothing, and omitting other
silently unmeters every non-Gemini model: the count is extracted and then
discarded, with no ratelimit reference anywhere in the transaction. Adding a
model is a feature-file edit, not a config change.
4. Deploy
Deploy to the service-extension env, then confirm the deployment reports
proxyDeploymentType: EXTENSIBLE.
curl -s -H "Authorization: Bearer $TOKEN" \
".../environments/$ENV/deployments"STANDARD means it will not govern the agent's off-Apigee tool egress.
If any policy calls a Google API — Model Armor, embeddings, Vector Search —
deploy with a service account or the deploy is rejected with
MISSING_SERVICE_ACCOUNT:
curl -X POST -H "Authorization: Bearer $TOKEN" \
".../apis/$PROXY/revisions/$REV/deployments?override=true&serviceAccount=$DEPLOY_SA"5. Networking
This is the gateway → Apigee leg only. PSC endpoint to the Apigee instance
service attachment, plus a private DNS A record. See
references/networking-psc.md. The Apigee instance consumerAcceptList must
include your project or the connection never reaches ACCEPTED.
The agent → gateway leg is the prerequisite above, and is not built here.
6. Bind the gateway
AuthzExtension then AuthzPolicy — in that order, and only once the host resolves and the proxy is deployed. The extension is inert until then.
Both document bodies ship as assets. Copy and substitute; do not invent them
— assets/authz-extension.example.yaml and assets/authz-policy.example.yaml.
They carry the fields that are load-bearing and have no obvious default:
wireFormat: EXT_PROC_GRPC, failOpen, the authority/service pair, the
metadata body flags, and the policy's customProvider / httpRules nesting.
gcloud service-extensions authz-extensions import $AUTHZ_EXT \
--source=authz-extension.yaml --location=$REGION --project=$PROJECT
gcloud beta network-security authz-policies import $AUTHZ_POLICY \
--source=authz-policy.yaml --location=$REGION --project=$PROJECTThe policy's resource paths take the project NUMBER, not the ID — both
target.resources and customProvider.authzExtension.resources. Resolve it
yourself; narrating two numbers at the user does not help. The trap, for when a
callout never lands: the number in the gateway's
serviceExtensionsServiceAccount is the Google-managed tenant project, not
theirs, and it is the only one on screen, so it is the natural thing to copy. A
wrong number imports cleanly and targets nothing. Copy the slots from
assets/authz-policy.example.yaml.
7. Verify
You cannot test with curl — the host is internal and the proxy expects the
gateway's IAP-egress JWT. Run scripts/read_debug_session.sh --arm, drive a
turn, read it back with --session. Expect one transaction per turn per
governed shape under a suffix list this narrow; zero alone does not mean the
callout is dead — the script prints the probe that settles it.
Hard rules
Gate every step — how urgently depends on your AuthzPolicy. An ungated
continueOnError="false" policy faults on any call that carries no prompt, and
that breaks the agent's turn. Historically the most common failure here.
A narrow suffix list routes only the model call to Apigee, so the trap mostly
does not fire. Broaden the policy and it fires at once: the ext-proc then sees
every agent egress — session create/read, appendEvent, OpenTelemetry exports —
measured at 12 calls for one single-question turn, of which one was the model
call. Gate the steps regardless; a condition is cheap and it is what makes the
policy safe to widen later. Full measurement: references/gotchas.md.
model call:
(request.uri Like "*generateContent*")
Never fail-close on your own metering. Keep continueOnError="true" on the
JavaScript and quota policies so a metering bug cannot break LLM traffic.
This skill meters by default and can enforce on request. Enforcement is
assets/feature-token-enforce.yaml: request-flow EnforceOnly policies sharing
each SharedName with the response-flow CountOnly half. Add it to the
template's feature list. Verified live on ext-proc traffic — the over-budget
caller gets a 429, model never called.
Never offer CountOnly=false as a substitute: in the response flow it cannot
gate (the answer is already on the wire) and it freezes the counter, losing the
metering too.
State two things whenever you offer it. It is one request late — the turn
that crosses the limit completes, because a token cost is not known until the
model answers. And it buckets on the requested model, which under an OpenAI
alias can differ from the resolved model the counting side bills. The two traps
that both present as "it never refuses": references/gotchas.md.
Normalize the model to a bucket. With an OpenAI-compatible endpoint the
model is an alias in the request body and the resolved name is in the response
body (google/gemini-2.5-pro). Exact-matching a raw name fails on the response
path.
jsonPath needs its third argument true so it returns a scalar. Without
it the quota policy receives a node it cannot parse.
References
references/modes.md— read first. How to answer "what can you do", dev vs prod, the defaults, durations, cost disclosure, the interlocks, and reading an assess reportscripts/assess_integration.sh— read-only. Setup Status Summary then a per-check table;--jsoncarries both. Lead with the summary and read out the Enforcement row. Run it before planning: it decides CREATE vs REUSE.TODOis a build step, not a failure; onlyFAILexits non-zeroreferences/networking-psc.md— both network legs, PSC, DNS, acceptancereferences/recipes.md— Model Armor and semantic cachereferences/verify.md— debug sessions and the checklist scriptreferences/setup.md— SDK, project, billing, APIs, region, auth, IAMreferences/gotchas.md— 63 failure modes and their symptomsassets/apigee-agent-gateway-governance.example.yaml— the input contractassets/apigee-proxy-template.yaml— the AFT template. Features merge bypriorityand all sit beside it:feature-identity.yaml(10),feature-normalize.yaml(15),feature-model-armor.yaml(20),feature-semantic-cache.yaml(30),feature-token-metering.yaml(50).assets/authz-extension.example.yaml,assets/authz-policy.example.yaml— the step-6 bodies. Copy and substitute; paths take the project numberAnything read out of the request body is normalised once, at priority 15, not inside whichever feature needs it first.
LLM_URI_MATCHis shared, so widening it arms every feature gated on it simultaneously — including Model Armor, which fails closed. Model Armor and the cache both readagw.user_prompt; the quotas readagw.total_tokens, set on the response path by the metering feature's own JS for the same reason.scripts/ensure_apigee_environment.sh— create the environment, attach it to the instance, create and attach the envgroup. Idempotent,--yesgatedscripts/ensure_gateway_connectivity.sh— route the gateway into the VPC. Read-modify-write;--yesgatedscripts/ensure_vector_search.sh— only whensemanticCacheis enabled: build index and endpoint, deploy one on the other, read back the domain.--yesgated, ~34 min, bills continuouslyscripts/teardown_vector_search.sh— the only thing that STOPS that charge. Refuses until the revision dropsSCL-lookup.--yesgatedscripts/deploy_proxy.sh— the Apigee-side sequence, idempotent,--dry-runscripts/read_debug_session.sh—--arma session (printing when it actually closes) and read it back as theverify.mdchecklist, instead of parsing the transaction JSON by handscripts/compile_check.py— compile the template offline, no org needed
Files
- ATTRIBUTION.mdAdded by AutoGineer880 B
- LICENSELicence file added by AutoGineer11.1 KiB
- SKILL.md26.1 KiB
- assets/apigee-agent-gateway-governance.example.yaml20.8 KiB
- assets/apigee-proxy-template.yaml9.4 KiB
- assets/authz-extension.example.yaml1.7 KiB
- assets/authz-policy.example.yaml2.1 KiB
- assets/feature-identity.yaml6.0 KiB
- assets/feature-model-armor.yaml3.5 KiB
- assets/feature-normalize.yaml4.0 KiB
- assets/feature-semantic-cache.yaml7.9 KiB
- assets/feature-token-enforce.yaml16.3 KiB
- assets/feature-token-metering.yaml14.9 KiB
- references/attribution.md1.9 KiB
- references/gotchas.md43.7 KiB
- references/modes.md46.5 KiB
- references/networking-psc.md10.0 KiB
- references/recipes.md9.4 KiB
- references/setup.md7.2 KiB
- references/verify.md14.3 KiB
- scripts/assess_integration.shScript35.5 KiB
- scripts/compile_check.pyScript9.9 KiB
- scripts/deploy_proxy.shScript23.7 KiB
- scripts/ensure_apigee_environment.shScript17.8 KiB
- scripts/ensure_gateway_connectivity.shScript21.2 KiB
- scripts/ensure_vector_search.shScript19.0 KiB
- scripts/read_debug_session.shScript15.5 KiB
- scripts/teardown_vector_search.shScript9.0 KiB
scripts/ensure_gateway_connectivity.sh
#!/bin/bash
# Give an Agent Gateway a route into your VPC: a PSC-interface network
# attachment plus DNS peering for the callout domain.
#
# Without both, the gateway cannot reach a private Apigee endpoint and cannot
# resolve the callout hostname. Nothing reports this. The gateway keeps serving
# ordinary agent traffic to public Google endpoints -- that path never enters
# your VPC -- so every other check passes and only the Apigee callout fails.
#
# Three things make this worth a script rather than three commands:
#
# 1. `agent-gateways import` is a WHOLE-RESOURCE upsert. Importing a YAML
# that carries only name + networkConfig silently drops `registries` and
# `protocols`. This does read-modify-write off the live export so nothing
# is lost.
# 2. `dnsPeeringConfig.targetNetwork` must be the full
# projects/{p}/global/networks/{n}. A bare network name is rejected.
# 3. A gateway created before `protocols` was enforced exports YAML that its
# own import rejects -- the API marks the field "Optional. Deprecated."
# while gcloud requires it client-side. PROTOCOL_UNSPECIFIED is injected
# when the export lacks it.
#
# It modifies a resource you own, so it applies nothing without --yes.
#
# Data goes to stdout as JSON. Diagnostics go to stderr.
set -euo pipefail
usage() {
cat <<'EOF'
Ensure an Agent Gateway can reach your VPC.
Usage:
ensure_gateway_connectivity.sh --gateway GW --project P --region R \
--vpc VPC --subnet SUBNET --attachment NAME --domain DOMAIN [options]
Required:
--gateway GW Agent Gateway short name
--project P project holding the gateway, VPC and attachment
--region R must match the gateway and the Apigee instance
--vpc VPC VPC the gateway will egress into
--subnet SUBNET subnet in --region for the network attachment
--attachment NAME network attachment; created if absent, reused if present
--domain DOMAIN private DNS domain to peer, trailing dot
(e.g. internal.example.local.)
Options:
--zone NAME private DNS zone to create for --domain when none exists.
The import is rejected outright if the domain has no
private zone in --vpc. Default: agent-gw-zone
--yes apply. Without it, prints the diff and exits 0.
--help this text
Environment:
CLOUDSDK_AUTH_ACCESS_TOKEN set from ADC if unset.
Exit codes:
0 already correct, or diff shown, or applied
1 usage error
2 precondition failed (missing subnet, gateway not found)
3 import failed or gateway still lacks networkConfig afterwards
EOF
}
log() { printf '%s\n' "$*" >&2; }
# ---------------------------------------------------------------- display ----
# Same palette as assess_integration.sh so the whole skill looks like one tool.
# The tty test is on fd 2, NOT fd 1: every log line here goes to stderr, so
# testing stdout would colour a piped run and leave an interactive one plain --
# exactly backwards. Honours NO_COLOR (no-color.org) and a dumb terminal.
if [[ -t 2 && -z "${NO_COLOR:-}" && "${TERM:-dumb}" != "dumb" ]]; then
C_RST=$'\033[0m'; C_DIM=$'\033[2m'; C_BLD=$'\033[1m'
C_GRN=$'\033[32m'; C_RED=$'\033[31m'; C_YEL=$'\033[33m'; C_BLU=$'\033[36m'
else
C_RST=; C_DIM=; C_BLD=; C_GRN=; C_RED=; C_YEL=; C_BLU=
fi
# hdr <title> section banner
# row <verb> <text> a planned change; CREATE/RESERVE cyan, MODIFY yellow
# cost <line>... standing-spend callout, always yellow
hdr() { printf '%s\n%s┌─ %s %s%s\n' "" "$C_DIM" "$1" "$(printf '─%.0s' $(seq 1 $(( 92 > ${#1} ? 92 - ${#1} : 3 ))))" "$C_RST" >&2; }
row() { local c="$C_BLU"; case "$1" in MODIFY|PATCH|APPEND) c="$C_YEL";; REUSE) c="$C_GRN";; esac
printf ' %s%-8s%s %s\n' "$c" "$1" "$C_RST" "$2" >&2; }
cost() { printf ' %s%s%s\n' "$C_YEL" "$*" "$C_RST" >&2; }
die() { log "ERROR: $1"; exit "${2:-1}"; }
GATEWAY=""; PROJECT=""; REGION=""; VPC=""; SUBNET=""; ATTACHMENT=""; DOMAIN=""; APPLY=0; ZONE="agent-gateway-callout-zone"
while [[ $# -gt 0 ]]; do
case "$1" in
--gateway) GATEWAY="$2"; shift 2 ;;
--project) PROJECT="$2"; shift 2 ;;
--region) REGION="$2"; shift 2 ;;
--vpc) VPC="$2"; shift 2 ;;
--subnet) SUBNET="$2"; shift 2 ;;
--attachment) ATTACHMENT="$2"; shift 2 ;;
--domain) DOMAIN="$2"; shift 2 ;;
--zone) ZONE="$2"; shift 2 ;;
--yes) APPLY=1; shift ;;
--help) usage; exit 0 ;;
*) usage >&2; die "unknown argument: $1" 1 ;;
esac
done
for v in GATEWAY PROJECT REGION VPC SUBNET ATTACHMENT DOMAIN; do
[[ -n "${!v}" ]] || { usage >&2; die "missing --${v,,}" 1; }
done
[[ "$DOMAIN" == *. ]] || die "--domain must end with a dot: $DOMAIN" 1
export CLOUDSDK_AUTH_ACCESS_TOKEN="${CLOUDSDK_AUTH_ACCESS_TOKEN:-$(gcloud auth application-default print-access-token)}"
WORK="$(mktemp -d)"; trap 'rm -rf "$WORK"' EXIT
# ---------------------------------------------------------------- inspect ----
log "==> reading gateway $GATEWAY"
gcloud network-services agent-gateways export "$GATEWAY" \
--location="$REGION" --project="$PROJECT" --destination="$WORK/before.yaml" \
>/dev/null 2>"$WORK/export.err" \
|| { cat "$WORK/export.err" >&2; die "cannot export gateway $GATEWAY" 2; }
# Surfacing an existing CUSTOM AuthzPolicy is informational. What is established:
# it cannot be merged with a new one -- `resources` is "Limited to 1 custom
# provider" and `policyProfile` is immutable -- so a second policy has to sit
# alongside it, or replace it.
#
# Deleting it is NOT known to break the gateway. On a live gateway the IAP
# CUSTOM policy was deleted during root-cause isolation, was not auto-recreated,
# and the gateway kept serving agent traffic with only the ext-proc policy
# attached. Whether it is safe in a given project depends on what relies on that
# policy, which this script cannot know -- so it reports and does not advise.
#
# Two CUSTOM policies coexisting is also NOT the cause of the 404-with-no-error
# failure; that is the environment's envgroup count. See references/gotchas.md,
# which lists a second CUSTOM policy as an individually disproved non-cause.
EXISTING_AUTHZ="$(gcloud network-security authz-policies list \
--location="$REGION" --project="$PROJECT" --format='value(name)' 2>/dev/null \
| while read -r pol; do
[[ -z "$pol" ]] && continue
if gcloud network-security authz-policies describe "${pol##*/}" \
--location="$REGION" --project="$PROJECT" --format='value(action,target.resources)' 2>/dev/null \
| grep -q "CUSTOM.*$GATEWAY"; then printf '%s ' "${pol##*/}"; fi
done)"
if [[ -n "$EXISTING_AUTHZ" ]]; then
log "NOTE: gateway $GATEWAY already has a CUSTOM AuthzPolicy: $EXISTING_AUTHZ"
log " It cannot be merged with a new one (one custom provider per policy,"
log " immutable policyProfile), so the ext-proc policy will sit alongside it."
log " Removing it may be safe -- check what depends on it first."
fi
# The two legs are INDEPENDENT and are detected separately. Testing for the
# `networkConfig:` block instead is the bug this comment exists to prevent: a
# gateway with egress attached but no DNS peering -- the exact state
# assess_integration.sh flags as
# dnsPeeringConfig absent -- ensure_gateway_connectivity.sh appends the domain
# -- took the early exit and reported "nothing to do". assess named this
# script as the remedy, this script declined to act, both exited 0, and the
# callout could not resolve its hostname while every later step passed.
HAS_EGRESS=0
grep -q 'networkAttachment:' "$WORK/before.yaml" && HAS_EGRESS=1
HAS_PEERING=0
grep -q 'dnsPeeringConfig:' "$WORK/before.yaml" && HAS_PEERING=1
# Whether the block exists is NOT the same question as whether OUR domain is in
# it. A gateway can already peer a different extension's domain -- the block is
# present, `domains` does not contain $DOMAIN, and the correct action is to
# APPEND to the list rather than to do nothing. Treating "block present" as
# "done" is the same class of bug as the one the comment above describes: the
# callout cannot resolve its hostname while every later step still passes.
DOMAIN_LISTED=0
if [[ $HAS_PEERING -eq 1 ]]; then
PRE_DOMAINS="$(gcloud network-services agent-gateways describe "$GATEWAY" \
--location="$REGION" --project="$PROJECT" \
--format='value(networkConfig.dnsPeeringConfig.domains)' 2>/dev/null || true)"
# gcloud joins a repeated field with ';'. Wrap both sides so the match is
# exact and a domain cannot match a longer one that merely contains it.
case ";$(printf '%s' "$PRE_DOMAINS" | tr -d '[:space:]');" in
*";$DOMAIN;"*) DOMAIN_LISTED=1 ;;
esac
fi
if [[ $HAS_EGRESS -eq 1 && $DOMAIN_LISTED -eq 1 ]]; then
log " egress attachment and DNS peering for $DOMAIN both present -- nothing to do"
printf '{"changed":false,"reason":"egress and dnsPeeringConfig both present","gateway":"%s"}\n' "$GATEWAY"
exit 0
fi
if [[ $HAS_EGRESS -eq 1 && $HAS_PEERING -eq 1 ]]; then
log " egress present, dnsPeeringConfig present but WITHOUT $DOMAIN -- appending"
elif [[ $HAS_EGRESS -eq 1 ]]; then
log " egress present, dnsPeeringConfig ABSENT -- gateway cannot resolve $DOMAIN"
elif [[ $HAS_PEERING -eq 1 ]]; then
log " dnsPeeringConfig present, egress ABSENT -- gateway has no route into $VPC"
else
log " networkConfig ABSENT -- gateway has no route into $VPC"
fi
gcloud compute networks subnets describe "$SUBNET" \
--region="$REGION" --project="$PROJECT" >/dev/null 2>&1 \
|| die "subnet $SUBNET not found in $REGION" 2
# The import validates dnsPeeringConfig.domains against EXISTING private zones in
# the target VPC and rejects the whole request if the domain has no zone:
# domain "x." does not exist as a private zone in network ".../networks/<vpc>"
# The zone may be empty -- only its existence is checked. The A record cannot be
# written yet anyway, since the PSC endpoint it points at comes later.
ZONE_EXISTS=0
EXISTING_ZONE="$(gcloud dns managed-zones list --project="$PROJECT" \
--filter="dnsName=$DOMAIN AND visibility=private" --format='value(name)' 2>/dev/null | head -1)"
if [[ -n "$EXISTING_ZONE" ]]; then
ZONE_EXISTS=1
ZONE="$EXISTING_ZONE"
fi
ATTACH_URI="projects/$PROJECT/regions/$REGION/networkAttachments/$ATTACHMENT"
ATTACH_EXISTS=0
gcloud compute network-attachments describe "$ATTACHMENT" \
--region="$REGION" --project="$PROJECT" >/dev/null 2>&1 && ATTACH_EXISTS=1
# ------------------------------------------------------------ build after ----
# Read-modify-write: start from the live export so registries, protocols and
# googleManaged survive. Append rather than regenerate.
cp "$WORK/before.yaml" "$WORK/after.yaml"
# Trap 3: a gateway predating client-side enforcement exports without
# `protocols`, and its own import then rejects it.
if ! grep -q '^protocols:' "$WORK/after.yaml"; then
log " export lacks 'protocols' -- injecting PROTOCOL_UNSPECIFIED so import is accepted"
printf 'protocols:\n- PROTOCOL_UNSPECIFIED\n' >>"$WORK/after.yaml"
fi
# Greenfield appends the whole block. When one leg already exists the block is
# there too, so the missing leg is INSERTED UNDER IT -- appending a second
# top-level `networkConfig:` would either be rejected or silently clobber the
# leg already in place, which for egress is the irreversible one.
#
# awk, not sed: the inserted text is multi-line and contains slashes, and the
# insert must happen exactly once even though `networkConfig:` appears in the
# comment-free export only at column 0.
_insert_under_networkconfig() {
awk -v block="$1" '
!done && /^networkConfig:[[:space:]]*$/ { print; printf "%s", block; done=1; next }
{ print }
END { if (!done) exit 9 }
' "$WORK/after.yaml" >"$WORK/after.yaml.tmp" \
|| die "export has no top-level 'networkConfig:' to insert under -- refusing to guess" 3
mv "$WORK/after.yaml.tmp" "$WORK/after.yaml"
}
# Adds one entry to an EXISTING dnsPeeringConfig.domains list, at the same
# indentation as the `domains:` key -- the shape PEERING_BLOCK below writes and
# the shape gcloud exports. Additive only: the entries already there are left
# untouched, because another extension depends on them.
_append_domain_to_peering() {
awk -v dom="$DOMAIN" '
/^[[:space:]]*domains:[[:space:]]*$/ && !done {
print
match($0, /^[[:space:]]*/)
print substr($0, 1, RLENGTH) "- " dom
done = 1
next
}
{ print }
END { if (!done) exit 9 }
' "$WORK/after.yaml" >"$WORK/after.yaml.tmp" \
|| die "dnsPeeringConfig present but has no 'domains:' list to append to -- refusing to guess" 3
mv "$WORK/after.yaml.tmp" "$WORK/after.yaml"
}
PEERING_BLOCK=" dnsPeeringConfig:
domains:
- $DOMAIN
targetNetwork: projects/$PROJECT/global/networks/$VPC
targetProject: $PROJECT
"
EGRESS_BLOCK=" egress:
networkAttachment: $ATTACH_URI
"
if [[ $HAS_EGRESS -eq 0 && $HAS_PEERING -eq 0 ]]; then
# printf, not a heredoc: both blocks already carry their trailing newline,
# and a heredoc terminator has to sit alone at the start of a source line.
printf 'networkConfig:\n%s%s' "$EGRESS_BLOCK" "$PEERING_BLOCK" >>"$WORK/after.yaml"
elif [[ $HAS_PEERING -eq 0 ]]; then
log " inserting dnsPeeringConfig into the existing networkConfig"
_insert_under_networkconfig "$PEERING_BLOCK"
else
# Peering block exists. Append our domain if it is missing -- never replace
# the list, which would unpeer whoever is already using this gateway.
if [[ $DOMAIN_LISTED -eq 0 ]]; then
log " appending $DOMAIN to the existing dnsPeeringConfig.domains"
_append_domain_to_peering
fi
if [[ $HAS_EGRESS -eq 0 ]]; then
log " inserting egress attachment into the existing networkConfig"
_insert_under_networkconfig "$EGRESS_BLOCK"
fi
fi
# --------------------------------------------------- egress reachability ----
# Attaching egress PINS ALL the gateway's egress into this VPC -- not just the
# Apigee callout. The agents' own control-plane calls (create_session, and the
# model call itself) start leaving through here too. If this VPC cannot reach
# Google APIs, every agent behind the gateway stops working.
#
# Checked BEFORE the diff, because the diff cannot show it: the change is
# purely additive on paper -- a networkConfig block appears, nothing is
# removed -- so it reads as safe while being a routing change. And it is NOT
# reversible (see the warning below), so a wrong answer here is expensive.
#
# ONLY when egress is actually being attached. If the gateway is already
# pinned and all we are adding is DNS peering, the pinning decision was taken
# long ago and this check can no longer protect anything -- it would just
# refuse the peering fix on a gateway whose callout is already broken, which
# is the worst moment to withhold the remedy.
SUBNET_PGA="n/a"
NAT_COUNT=0
if [[ $HAS_EGRESS -eq 0 ]]; then
SUBNET_PGA="$(gcloud compute networks subnets describe "$SUBNET" \
--region="$REGION" --project="$PROJECT" \
--format='value(privateIpGoogleAccess)' 2>/dev/null || echo "")"
NAT_COUNT="$(gcloud compute routers list --project="$PROJECT" \
--filter="region:($REGION) AND network:($VPC)" \
--format='value(name)' 2>/dev/null | grep -c . || true)"
if [[ "$SUBNET_PGA" != "True" && "${NAT_COUNT:-0}" -eq 0 ]]; then
log ""
log "!! REFUSING: subnet '$SUBNET' has privateIpGoogleAccess=${SUBNET_PGA:-unset} and"
log "!! there is no Cloud NAT on network '$VPC' in $REGION."
log "!!"
log "!! Pinning gateway egress into a VPC that cannot reach *.googleapis.com"
log "!! takes EVERY agent behind this gateway offline, not just the Apigee"
log "!! callout -- the agents' own create_session calls egress here too. The"
log "!! symptom is an agent-side connection reset to"
log "!! <region>-aiplatform.mtls.googleapis.com, which reads as an agent bug."
log "!!"
log "!! Fix one of these first, then re-run:"
log "!! gcloud compute networks subnets update $SUBNET --region=$REGION \\"
log "!! --project=$PROJECT --enable-private-ip-google-access"
log "!! (or create a Cloud NAT on $VPC in $REGION)"
die "VPC egress cannot reach Google APIs; refusing to pin gateway egress into it" 2
fi
log " egress reachability: privateIpGoogleAccess=$SUBNET_PGA, CloudNAT routers=$NAT_COUNT -- OK"
else
log " egress already attached -- reachability check skipped, pinning is not being changed"
fi
# ------------------------------------------------------------------ diff -----
# The one-way warning belongs to the EGRESS leg only. Printing it when all we
# are adding is DNS peering trains the reader to scroll past the loudest
# warning in the skill, which is how it stops working the day it matters.
log ""
if [[ $HAS_EGRESS -eq 0 ]]; then
log "!! ONE-WAY CHANGE. egress.networkAttachment is settable once and IMMUTABLE"
log "!! afterwards -- removing it is itself a mutation and is rejected:"
log "!! \"network attachment is not mutable\""
log "!! The attachment cannot be deleted while the gateway holds it either"
log "!! (\"Network Attachment with connected endpoints cannot be deleted\")."
log "!! The only way back is deleting and recreating the Agent Gateway, which"
log "!! means rebinding every agent -- and agent_gateway_config is set at agent"
log "!! CREATION time, so in practice the agents get redeployed."
log "!! This is the only script here that changes a resource you already own"
log "!! in a way you cannot undo."
log ""
fi
hdr "Proposed change to $GATEWAY"
if [[ $ZONE_EXISTS -eq 1 ]]; then
row REUSE "private DNS zone $ZONE for $DOMAIN"
else
row CREATE "private DNS zone $ZONE for $DOMAIN -- the import is rejected without it"
fi
if [[ $HAS_EGRESS -eq 1 ]]; then
row REUSE "network attachment (gateway egress already attached)"
elif [[ $ATTACH_EXISTS -eq 1 ]]; then
row REUSE "network attachment $ATTACHMENT"
else
row CREATE "network attachment $ATTACHMENT"
fi
if [[ $HAS_EGRESS -eq 1 ]]; then
row MODIFY "$GATEWAY -- DNS peering for $DOMAIN only (diff below)"
elif [[ $HAS_PEERING -eq 1 ]]; then
row MODIFY "$GATEWAY -- egress attachment only (diff below)"
else
row MODIFY "$GATEWAY -- egress attachment and DNS peering (diff below)"
fi
log ""
diff -u "$WORK/before.yaml" "$WORK/after.yaml" >&2 || true
log ""
# The single most consequential warning in the skill, and it used to be in
# prose elsewhere rather than at the point of consent. Gated with the leg it
# describes -- see the note above the diff.
if [[ $HAS_EGRESS -eq 0 ]]; then
cost "IRREVERSIBLE egress.networkAttachment is immutable once set, and it pins"
cost " ALL of this gateway's egress into that VPC -- not just the"
cost " Apigee callout. Agents already running can lose access to"
cost " Google APIs unless the subnet has Private Google Access or"
cost " a Cloud NAT. Capture a known-good agent turn before applying."
log ""
fi
if [[ $APPLY -ne 1 ]]; then
printf ' %sNot applying.%s Re-run with %s--yes%s to make these changes.\n' \
"$C_BLD" "$C_RST" "$C_BLD" "$C_RST" >&2
printf '{"changed":false,"reason":"dry run, --yes not given","gateway":"%s","attachmentExists":%s}\n' \
"$GATEWAY" "$([[ $ATTACH_EXISTS -eq 1 ]] && echo true || echo false)"
exit 0
fi
# ----------------------------------------------------------------- apply -----
if [[ $ZONE_EXISTS -eq 0 ]]; then
log "==> creating private DNS zone $ZONE for $DOMAIN"
gcloud dns managed-zones create "$ZONE" \
--dns-name="$DOMAIN" --visibility=private \
--networks="$VPC" --description="Agent Gateway ext-proc callout zone" \
--project="$PROJECT" >/dev/null
fi
if [[ $ATTACH_EXISTS -eq 0 ]]; then
log "==> creating network attachment $ATTACHMENT"
gcloud compute network-attachments create "$ATTACHMENT" \
--region="$REGION" --subnets="$SUBNET" \
--connection-preference=ACCEPT_AUTOMATIC \
--project="$PROJECT" >/dev/null
fi
# The export carries an etag, so a concurrent edit fails the import instead of
# being silently overwritten.
log "==> importing gateway with networkConfig"
gcloud network-services agent-gateways import "$GATEWAY" \
--source="$WORK/after.yaml" --location="$REGION" --project="$PROJECT" \
>/dev/null 2>"$WORK/import.err" \
|| { cat "$WORK/import.err" >&2; die "import failed -- gateway unchanged" 3; }
# Verify the leg that was actually added, not just that networkConfig exists.
# Checking only egress passes trivially in the peering-only case -- egress was
# already there before the import -- so a peering insert that silently did
# nothing would still be reported as success. That is the same shape of
# false-green this script was just fixed for.
log "==> verifying"
ACTUAL="$(gcloud network-services agent-gateways describe "$GATEWAY" \
--location="$REGION" --project="$PROJECT" \
--format='value(networkConfig.egress.networkAttachment)' 2>/dev/null || true)"
[[ -n "$ACTUAL" ]] || die "import reported success but egress attachment is still absent" 3
log " egress attachment: $ACTUAL"
PEER_DOMAINS="$(gcloud network-services agent-gateways describe "$GATEWAY" \
--location="$REGION" --project="$PROJECT" \
--format='value(networkConfig.dnsPeeringConfig.domains)' 2>/dev/null || true)"
case "$PEER_DOMAINS" in
*"$DOMAIN"*) log " dns peering: $PEER_DOMAINS" ;;
*) die "import reported success but dnsPeeringConfig does not list $DOMAIN (got: ${PEER_DOMAINS:-none})" 3 ;;
esac
printf '{"changed":true,"gateway":"%s","networkAttachment":"%s","dnsDomain":"%s","dnsPeeringDomains":"%s"}\n' \
"$GATEWAY" "$ACTUAL" "$DOMAIN" "$PEER_DOMAINS"
Requirements and compatibility
Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Kiro, OpenCode, Goose, Amp, Roo Code, JetBrains Junie, Factory Droid, Devin (ex-Windsurf), Mistral Vibe, OpenHands, Letta
“Works with” means the tool reads the open Agent Skills format; a skill written for one tool may use features others lack.
Needs: Needs a command-line tool, Needs Python, Needs Node.js
Security
What our scanner noticed· Low risk
- Instructions that use the network
- Scripts that your agent may run
This skill includes scripts that your agent may run. Read them before you use it.
A static scan is not a guarantee. Skills are third-party content: review a skill before you let an agent use it.
Attribution and licence
- Published by
- Publisher
- Google (Official)
- Source
- google/skills · skills/cloud/apigee-agent-gateway-governance · commit 2194d3b
- Licence
- Apache License 2.0 (redistribution allowed)
- LICENSE (added by AutoGineer from LICENSE)
- ATTRIBUTION.md (written by AutoGineer)
- Synced
AutoGineer is not affiliated with or endorsed by Google or Google.
Download ZIP
The ZIP holds the skill’s files, its licence files and an ATTRIBUTION.md. Download apigee-agent-gateway-governance.zip