Update the Agent Platform skills.

- Remove the internal sandbox endpoint override from the deploy skill's gcloud setup script, so it configures the default Vertex AI endpoint.
- Stop each skill from building its own virtual environment; install only what is missing.  Each skill used to create a virtualenv and install its own SDKs before running anything. Because a virtualenv is isolated from the surrounding interpreter, that discarded SDKs which were already present and made every skill pay for a redundant install. The skills now check whether their imports resolve and install only what is actually missing, so the common case does no work at all.  requirements.txt is unchanged in purpose: it remains the manifest for anyone who downloads a single skill on its own. It should not be applied on top of a working environment, where its pins would downgrade SDKs that other skills share. Its numpy, pandas and pyarrow pins are dropped, as no script in the skill imports them.  calculate_cost.py no longer depends on smart_open. It reads local paths with the builtin open() and gs:// paths with google-cloud-storage, which the skill already has. This removes the one dependency that could not be satisfied by an ordinary unpinned install, because no single smart_open release exposed the API used here across the supported range.  Also fixes two latent bugs in the setup instructions: a pip version specifier was unquoted, so the shell parsed ">=1.154.0" as a redirect and wrote an empty file instead of constraining the install; and one skill installed the separate, deprecated vertexai distribution, which already ships inside google-cloud-aiplatform.
- Pin google-genai >=1.74.0 in the tuning skill so enterprise=True works.  The tuning skill pinned google-genai==1.73.1, but tune_open_model.py and monitor_tuning_job.py call genai.Client(enterprise=True, ...), which requires google-genai 1.74.0 or later, so those scripts raised a TypeError. Bump the pin to 1.75.0, the highest release that still satisfies the google-genai<2.0.0 constraint declared by google-cloud-aiplatform 1.138.0, and add a test that keeps the pin in range.
- Add Qwen 3.5 9B to the open model tuning catalog, and tell the agent to offer the closest supported model when a user asks for one that is not supported for tuning.
- Remove stray internal lint-suppression comments from the published skills, and fix a malformed code block in the tuning skill. The disableFinding and enableFinding HTML comments were being copied verbatim into the published files instead of being stripped, and are now removed. The upload snippet in Phase 1.2 of the tuning skill rendered as a broken inline code span and is now a proper bash code block.
- Internal change
- Recommend the `global` location for open model tuning. The skill previously applied one five-region whitelist to every tuning job and rejected `global` outright, which refused valid open model requests and forced a region choice before any other setup. It now recommends `global` for open models and confirms it with the user, documents when an explicit region is still required (CMEK and data residency), and notes that sub-resources resolve to a real region that must be read before monitoring or deployment. For Gemini, supported regions vary per model, so the skill directs a per-model documentation lookup instead of reusing the open model list.
- [agent_platform] Narrow the `agent-platform-deploy` skill trigger description so that pure listing/discovery questions ("is X deployed?", "list my endpoints", "which regions have models running?") route to `agent-platform-endpoint-management` instead of speculatively loading the deploy skill. Also clarify that "list available Model Garden models" refers to the model CATALOG (not the user's endpoints), and replace "check deployment status, verify serving endpoints" with the more specific "check the status of an in-progress deployment operation".

PiperOrigin-RevId: 957218689

Author: cloud-ix-copybara

skills/cloud/agent-platform-deploy/SKILL.md

@@ -4,14 +4,19 @@ metadata:
   category: AiAndMachineLearning
 description: >-
   Deploy open models or custom weights from Model Garden to Agent Platform
-  endpoints, check deployment status, verify serving endpoints, or clean up
-  resources by undeploying models and deleting endpoints. Use when asked to
-  deploy models on Agent Platform, list available Model Garden models, check if
-  a model is deployable, query deployment cost, troubleshoot deployment errors
-  (like quota limits), or undeploy/clean up endpoints. Also use when copying
-  and deploying a 1P Tuned Model. Don't use for public Vertex AI deployments
-  (use the `vertex-deploy` skill) or for running model evaluations (use the
-  `agent-platform-eval-flywheel` skill).
+  endpoints, check the status of an in-progress deployment operation, or clean
+  up resources by undeploying models and deleting endpoints. Use when asked to
+  actively deploy a model, list the Model Garden CATALOG of available models,
+  check if a specific model is deployable
+  (`gcloud ai model-garden models list-deployment-config`), query deployment
+  cost, troubleshoot deployment errors (like quota limits), or undeploy/clean
+  up endpoints. Also use when copying and deploying a 1P Tuned Model. Don't
+  use for pure listing/discovery questions of the form "is X deployed?",
+  "list my endpoints", or "which regions have models running?" — for those
+  use `agent-platform-endpoint-management`. Don't use for public Vertex AI
+  deployments (use the `vertex-deploy`
+  skill) or for running model evaluations (use the `agent-platform-eval-flywheel`
+  skill).
 ---
 
 # Agent Platform Model Garden Deploy Skill
@@ -188,6 +193,7 @@ Interpret the probe result and act:
     to this project's Vertex AI API — enable it in the console or switch
     projects") and the concrete next action.
 
+
 ## 3. Deploying a Model
 
 > [!WARNING] Deploying models, especially large ones, consumes significant

skills/cloud/agent-platform-deploy/references/copy_deploy_guide.md

@@ -46,21 +46,22 @@ If user is copying model in different region, skip the P4SA setup section.
 -   Verify `gcloud auth list`. If not authenticated, run `gcloud auth login`.
 -   Execute the following command to set the global variable.
     `export PROJECT_ID=${PROJECT_ID} REGION=${REGION}`
--   Check if ${USER} have value, or ask user to set one.
+-   Check if ${USER_EMAIL} have value (the full account email), or ask user to
+    set one.
 
 ### 0.2 GCloud CLI setup
 
--   use `scripts/config_gcloud_cli.sh ${ENV} ${PROJECT_ID} ${REGION} ${USER}`
+-   use `scripts/config_gcloud_cli.sh ${ENV} ${PROJECT_ID} ${REGION} ${USER_EMAIL}`
 
 ### 0.3 P4SA Setup
 
 #### 0.3.0 Goal
 
 To copy a model from source project ${SOURCE_PROJECT} to the destination project
 ${PROJECT_ID}, and ${REGION}, follow
-<!-- disableFinding(LINE_OVER_80) -->
+
 https://docs.cloud.google.com/gemini-enterprise-agent-platform/machine-learning/model-registry/copy-model, add the
-<!-- enableFinding(LINE_OVER_80) -->
+
 P4SA of the destination project as a new principal to the source project and
 assign the Vertex AI Service Agent role to it.
 

skills/cloud/agent-platform-deploy/scripts/config_gcloud_cli.sh

@@ -3,15 +3,15 @@
 ENV=${1:-$ENV}
 PROJECT_ID=${2:-$PROJECT_ID}
 REGION=${3:-$REGION}
-USER=${4:-$USER}
+USER_EMAIL=${4:-$USER_EMAIL}
 
 if [[ -z "${PROJECT_ID}" ]]; then
     echo "Error: PROJECT_ID is not set (neither as an argument nor as an environment variable)."
     exit 1
 fi
 
-if [[ -z "${USER}" ]]; then
-    echo "Error: USER is not set (neither as an argument nor as an environment variable)."
+if [[ -z "${USER_EMAIL}" ]]; then
+    echo "Error: USER_EMAIL is not set (neither as an argument nor as an environment variable)."
     exit 1
 fi
 
@@ -24,20 +24,17 @@ if [[ -z "${REGION}" ]]; then
     exit 1
 fi
 
-ENDPOINT="https://${REGION}-${ENV}-aiplatform.sandbox.googleapis.com"
 
 echo "PROJECT_ID: ${PROJECT_ID}"
-echo "USER: ${USER}"
+echo "Account: ${USER_EMAIL}"
 echo "Env: ${ENV}"
 echo "Region: ${REGION}"
-echo "Endpoint: ${ENDPOINT}"
 
 if ! gcloud config configurations describe "${ENV}-cdmodel" > /dev/null 2>&1; then
   gcloud config configurations create "${ENV}-cdmodel"
   gcloud config set core/project "${PROJECT_ID}"
   gcloud config set compute/region "${REGION}"
-  gcloud config set account "${USER}"@google.com
-  gcloud config set api_endpoint_overrides/aiplatform "${ENDPOINT}"
+  gcloud config set account "${USER_EMAIL}"
 fi
 
 gcloud config configurations activate ${ENV}-cdmodel

skills/cloud/agent-platform-endpoint-management/SKILL.md

@@ -74,6 +74,10 @@ gcloud ai endpoints list \
     --region=$LOCATION_ID
 ```
 
+*(Optional)* For pagination, you MUST use `--limit=$LIMIT` to restrict the total
+number of returned endpoints. You can also append `--page-size=$PAGE_SIZE` to
+control API chunking, or `--page-token=$PAGE_TOKEN` for next pages.
+
 > [!IMPORTANT]
 >
 > Always specify the `--region`. Do NOT use 'global'. Ask the user to specify if

skills/cloud/agent-platform-eval-flywheel/SKILL.md

@@ -53,12 +53,19 @@ to the following safety tiers based on the action requested:
 
 ## Setup
 
-Install the SDK:
+The scripts need `vertexai` (from `google-cloud-aiplatform[evaluation]`),
+`google-genai`, `pandas`, and `requests`. Do **not** create a virtual
+environment — it starts empty and hides packages the environment already
+provides, forcing a redundant install. Probe, and install only what is missing:
 
 ```bash
-pip install google-cloud-aiplatform[evaluation]>=1.154.0 google-genai>=1.0.0
+python3 -c "import vertexai, google.genai, pandas, requests" \
+  || pip install 'google-cloud-aiplatform[evaluation]>=1.154.0' 'google-genai>=1.0.0'
 ```
 
+The version specifiers must stay quoted: unquoted, bash reads `>=1.154.0` as a
+redirect and silently writes an empty file instead of constraining the install.
+
 Need `GOOGLE_CLOUD_PROJECT` and `GOOGLE_CLOUD_LOCATION`. Check env vars first;
 if missing, ask the user. Newer Gemini models often need `location="global"`.
 

skills/cloud/agent-platform-eval-flywheel/references/deployment.md

@@ -31,15 +31,19 @@ only covers the deployment + endpoint-inference subset.
 
 ## Setup
 
-Install the deployment-eval dependencies (in addition to the base SDK install in
-the main SKILL.md):
+The deployment-eval dependencies are the same set the main SKILL.md relies on.
+Do **not** create a virtual environment — it starts empty and hides packages the
+environment already provides, forcing a redundant install. Probe, and install
+only what is missing:
 
 ```bash
-python3 -m venv .model-eval
-source .model-eval/bin/activate
-pip install google-cloud-aiplatform[evaluation]>=1.154.0 google-genai>=1.0.0 requests
+python3 -c "import vertexai, google.genai, requests" \
+  || pip install 'google-cloud-aiplatform[evaluation]>=1.154.0' 'google-genai>=1.0.0' requests
 ```
 
+Keep the version specifiers quoted: unquoted, bash reads `>=1.154.0` as a
+redirect and silently writes an empty file instead of constraining the install.
+
 The scripts also shell out to `gcloud` and `gsutil`, so the user must have the
 Google Cloud SDK installed and `gcloud auth application-default login`
 completed.

skills/cloud/agent-platform-inference/SKILL.md

@@ -71,32 +71,32 @@ environment is correctly initialized by following these steps:
     gcloud services enable aiplatform.googleapis.com
     ```
 
-3.  **Virtual Environment**: Create and activate a dedicated local virtual
-    environment:
+3.  **Python Dependencies**: The scripts import `vertexai` (from
+    `google-cloud-aiplatform`), `google-genai`, and `openai`. Do **not** create
+    a virtual environment — it starts empty and hides packages the environment
+    already provides, forcing a redundant install. Probe, and install only what
+    is missing:
 
     ```bash
-    python3 -m venv .venv
-    source .venv/bin/activate
+    python3 -c "import vertexai, google.genai, openai" \
+      || pip install -r scripts/requirements.txt
     ```
 
-4.  **Install Dependencies**: Install the required SDKs:
+    The pins in `scripts/requirements.txt` are a fallback for an environment
+    that does not already provide these SDKs; do not apply them on top of a
+    working environment.
 
-    ```bash
-    pip install -r scripts/requirements.txt
-    ```
-
-5.  **Verify Setup (Optional)**: Run all sample scripts at once to verify the
+4.  **Verify Setup (Optional)**: Run all sample scripts at once to verify the
     environment is working end-to-end:
 
     ```bash
     ./scripts/verify_all.sh
     ```
 
-6.  **Execution**: Advise the user that every time they execute a Python snippet
-    from this skill, they must ensure this virtual environment is activated
-    first.
+5.  **Execution**: Run the scripts with a plain `python3 scripts/...`. There is
+    no environment to activate first.
+
 
-<!-- disableFinding(LINE_OVER_80) -->
 
 > [!IMPORTANT] **CRITICAL: Model IDs & Availability** * **Gemini Models**: See
 > [Gemini Models][gemini-models-docs] for valid Model IDs and Regions. *
@@ -109,7 +109,7 @@ environment is correctly initialized by following these steps:
 >
 > \[gemini-models-docs]:
 > https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/migrate
-> <!-- enableFinding(LINE_OVER_80) -->
+>
 >
 > ## Workflow Decision Tree
 
@@ -288,21 +288,21 @@ def get_gcp_access_token():
 
 > [!NOTE] Google Cloud access tokens typically expire after 1 hour. The
 > `get_gcp_access_token()` function above retrieves a *fresh* token at the time
-> it is called. <!-- disableFinding(LINE_OVER_80) --> For long-running
+> it is called. For long-running
 > applications, you implement a refresh mechanism. See
 > [Refresh the access token](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/migrate/openai/auth-and-credentials?hl=en#refresh_your_credentials)
-> for details. <!-- enableFinding(LINE_OVER_80) -->
+> for details.
 
 ### Configuration (Base URL)
 
-<!-- disableFinding(LINE_OVER_80) -->
+
 
 -   **Global Endpoint** (Recommended for most models requiring global
     availability):
     `https://aiplatform.googleapis.com/v1/projects/{PROJECT_ID}/locations/global/endpoints/openapi`
 -   **Regional Endpoint**:
     `https://{REGION}-aiplatform.googleapis.com/v1/projects/{PROJECT_ID}/locations/{REGION}/endpoints/openapi`
-    <!-- enableFinding(LINE_OVER_80) -->
+
 
 ### Python Example (OpenMaaS - Chat Completions)
 
@@ -708,10 +708,10 @@ print(response.text)
     1.  **Check Location Availability**:
         *   **OpenMaaS**: Verify the model is available in your region. See
             [Model Availability by Location](https://docs.cloud.google.com/gemini-enterprise-agent-platform/resources/locations#genai-open-models).
-        *   **Gemini**: <!-- disableFinding(LINE_OVER_80) -->
+        *   **Gemini**:
             *   **Source of Truth**: Always check
                 [Gemini Model Locations](https://docs.cloud.google.com/gemini-enterprise-agent-platform/resources/locations#google-models)
-                for the authoritative list. <!-- enableFinding(LINE_OVER_80) -->
+                for the authoritative list.
             *   **Preview Models**: All Preview models (e.g., Gemini 3.1,
                 experimental versions) are often **ONLY** available in the
                 `us-central1` or `global` regions.

skills/cloud/agent-platform-inference/scripts/verify_all.sh

@@ -1,21 +1,12 @@
 #!/bin/bash
 set -e
 
-# Create a temporary directory for the virtual environment
-VENV_DIR=$(mktemp -d -t venv_verify.XXXXXX)
-python3 -m venv "$VENV_DIR"
-source "$VENV_DIR/bin/activate"
-
-# Trap to ensure cleanup happens on exit
-cleanup() {
-  echo "Cleaning up virtual environment..."
-  deactivate 2>/dev/null || true
-  rm -rf "$VENV_DIR"
-}
-trap cleanup EXIT
-
-echo "Installing requirements..."
-pip install -q -r scripts/requirements.txt
+# Verify against the ambient interpreter, which is the one the skill's scripts
+# actually run under. A throwaway venv would verify an environment no real
+# invocation ever uses, so it could pass while the real one is broken.
+echo "Checking dependencies..."
+python3 -c "import vertexai, google.genai, openai" \
+  || pip install -q -r scripts/requirements.txt
 
 echo "Running verification tests..."
 FAILED=0

skills/cloud/agent-platform-prompt-management/SKILL.md

@@ -83,22 +83,18 @@ these steps:
     gcloud auth application-default login
     ```
 
-2.  **Virtual Environment**: Create and activate a dedicated virtual
-    environment:
+2.  **Python Dependencies**: This skill needs `google-cloud-aiplatform` and
+    `google-genai`. Do **not** create a virtual environment — it starts empty
+    and hides packages the environment already provides, forcing a redundant
+    install. Probe, and install only what is missing:
 
     ```bash
-    python3 -m venv ~/prompt_agent_venv
-    source ~/prompt_agent_venv/bin/activate
+    python3 -c "import vertexai, google.genai" \
+      || pip install google-cloud-aiplatform google-genai
     ```
 
-3.  **Install Dependencies**: Install the required Agent Platform SDKs:
-
-    ```bash
-    pip install google-cloud-aiplatform google-genai
-    ```
-
-4.  **Execution**: Advise the user that every time they execute a Python
-    snippet, they must ensure this virtual environment is activated first.
+3.  **Execution**: Run Python snippets with a plain `python3`. There is no
+    environment to activate first.
 
 > [!TIP]
 >

skills/cloud/agent-platform-tuning-management/SKILL.md

@@ -37,14 +37,7 @@ following safety tiers based on the action requested:
 **CRITICAL**: Before running any of the Python snippets below, you MUST ensure
 the environment is correctly initialized by following these steps:
 
-1.  **Virtual Environment**: Create and activate a virtual environment:
-
-    ```bash
-    python3 -m venv ~/tuning_mgr_venv
-    source ~/tuning_mgr_venv/bin/activate
-    ```
-
-2.  **Google Cloud Authentication**: Authenticate with your Google Cloud account
+1.  **Google Cloud Authentication**: Authenticate with your Google Cloud account
     and configure active Application Default Credentials (ADC) for Agent
     Platform access:
 
@@ -53,14 +46,17 @@ the environment is correctly initialized by following these steps:
     gcloud auth application-default login
     ```
 
-3.  **Install Dependencies**: Install the required Agent Platform SDK:
+2.  **Python Dependencies**: This skill needs `google-cloud-aiplatform`. Do
+    **not** create a virtual environment — it starts empty and hides packages
+    the environment already provides, forcing a redundant install. Probe, and
+    install only what is missing:
 
     ```bash
-    pip install google-cloud-aiplatform
+    python3 -c "import vertexai" || pip install google-cloud-aiplatform
     ```
 
-4.  **Execution**: Advise the user that every time they execute a Python
-    snippet, they must ensure this virtual environment is activated first.
+3.  **Execution**: Run Python snippets with a plain `python3`. There is no
+    environment to activate first.
 
 ## Workflow Decision Tree