Index of every page on this site: llms.txt
All skills
jeremylongshore avatar

/adk-deployment-specialist

@820dfc8
Run it yourself

Deploy and orchestrate Vertex AI ADK agents using A2A protocol. Manages AgentCard discovery, task submission, Code Execution Sandbox, and Memory Bank. Use when asked to "deploy ADK agent" or "orchestrate agents". Trigger with phrases like 'deploy', 'infrastructure', or 'CI/CD'.

Use this Skill: https://skilld.dev/gh/jeremylongshore/tons-of-skills-marketplace/adk-deployment-specialist

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

referenceserrors.md

≈1.5k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Error Handling Reference

Deployment Failures

google.api_core.exceptions.NotFound: 404 Reasoning engine not found

  • Cause: The reasoning engine resource name is incorrect or the resource was deleted.
  • Fix: Verify the resource name with client.agent_engines.list() and confirm the project/location match.

google.api_core.exceptions.PermissionDenied: 403 Permission denied on resource

  • Cause: The service account or user lacks required IAM roles.
  • Fix: Grant roles/aiplatform.user for querying agents, roles/aiplatform.admin for creating/deleting. Use gcloud projects get-iam-policy PROJECT_ID to audit current bindings.

google.api_core.exceptions.InvalidArgument: 400 Invalid agent configuration

  • Cause: The agent definition has invalid fields, unsupported model, or malformed tools.
  • Fix: Verify the model string (e.g., gemini-2.5-flash), ensure all tool functions have proper type hints and docstrings, and check that requirements list includes all needed packages.

google.api_core.exceptions.ResourceExhausted: 429 Quota exceeded

  • Cause: Project has hit the Agent Engine quota (default: 10 reasoning engines per project).
  • Fix: Request quota increase via Cloud Console > IAM & Admin > Quotas, or delete unused agents with client.agent_engines.delete(name=resource_name).

Deployment hangs or times out

  • Cause: Large dependency list, network issues, or package build failures in the remote environment.
  • Fix: Minimize requirements to only what the agent needs. Pin exact versions. Check Cloud Build logs via gcloud builds list --project=PROJECT_ID.

Authentication & Authorization Issues

google.auth.exceptions.DefaultCredentialsError: Could not automatically determine credentials

  • Cause: No Application Default Credentials configured.
  • Fix: Run gcloud auth application-default login for local development, or set GOOGLE_APPLICATION_CREDENTIALS to a service account key path. For production, use Workload Identity Federation.

google.auth.exceptions.RefreshError: ('invalid_grant: Token has been expired or revoked')

  • Cause: Cached credentials are stale or the service account key was rotated.
  • Fix: Re-authenticate with gcloud auth application-default login. For service accounts, generate a new key and update GOOGLE_APPLICATION_CREDENTIALS.

Agent runs but returns permission errors at runtime

  • Cause: The Agent Engine service agent (the identity the deployed agent runs as) lacks permissions to call downstream APIs (e.g., GKE, Cloud Run).
  • Fix: Grant the Agent Engine service agent (service-PROJECT_NUMBER@gcp-sa-aiplatform-re.iam.gserviceaccount.com) the required roles for the APIs the agent's tools invoke.

Orchestration Failures

Sub-agent in SequentialAgent fails mid-pipeline

  • Cause: One agent in the chain returned an error or produced output the next agent could not process.
  • Fix: Wrap each sub-agent in error-handling instructions. Add a try/except around the orchestrator call. Check which sub-agent failed by inspecting the events in the runner response.

ParallelAgent produces inconsistent results

  • Cause: Parallel sub-agents share no state by default; if they depend on each other's output, results are non-deterministic.
  • Fix: Use SequentialAgent when ordering matters. Only use ParallelAgent for truly independent tasks.

LoopAgent exceeds max iterations

  • Cause: The exit condition was never met, or the loop agent's should_continue logic has a bug.
  • Fix: Set max_iterations on the LoopAgent. Add explicit exit conditions in the agent's instructions. Monitor iteration count in logs.

SDK & Import Errors

ImportError: cannot import name 'Agent' from 'google.adk'

  • Cause: Using the wrong import path.
  • Fix: The correct import is from google.adk.agents import Agent. Other common imports:
    • from google.adk.agents import SequentialAgent, ParallelAgent, LoopAgent
    • from google.adk.runners import Runner
    • from google.adk.sessions import VertexAiSessionService
    • from google.adk.tools import FunctionTool

ModuleNotFoundError: No module named 'google.adk'

  • Cause: ADK SDK not installed or wrong Python environment.
  • Fix: Install with pip install google-adk>=1.15.1. Verify with python -c "import google.adk; print(google.adk.__version__)".

Agent Engine Management (No gcloud CLI)

There is no gcloud ai agents or gcloud alpha ai agent-engines CLI. All Agent Engine management is done via the Python SDK:

import vertexai

client = vertexai.Client(project="PROJECT_ID", location="us-central1")

# List agents
agents = client.agent_engines.list()

# Get a specific agent
agent = client.agent_engines.get(
    name="projects/PROJECT/locations/LOC/reasoningEngines/ID"
)

# Create/deploy an agent
remote = client.agent_engines.create(agent_engine=my_agent, requirements=[...])

# Delete an agent
client.agent_engines.delete(name=agent.resource_name)

A2A Protocol Errors

AgentCard not found at /.well-known/agent-card

  • Cause: The agent does not expose an A2A-compliant AgentCard, or the URL is wrong.
  • Fix: Verify the agent was deployed with A2A support. The AgentCard endpoint is /.well-known/agent-card (note: no trailing slash). Ensure the agent framework generates this endpoint.

tasks/send returns 405 Method Not Allowed

  • Cause: The agent endpoint does not support the A2A task submission method.
  • Fix: Confirm the agent was built with A2A protocol support (a2a-sdk). Check that the server handles POST to /tasks/send with JSON-RPC 2.0 format.

Task polling returns stale status

  • Cause: Eventual consistency or caching in the agent's task store.
  • Fix: Implement exponential backoff in polling (start at 2s, max 30s). Use the task's updatedAt timestamp to detect stale responses.

Tons of Skills by Intent Solutions | jeremylongshore.com

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

2 warnings2mo3 checks · Risk MEDIUM
  • Gen Agent Trust Hub2mo

    The skill is designed for deploying Google Vertex AI agents and includes scripts for deployment and protocol testing. While generally functional, it contains a testing utility that retrieves sensitive GCP access tokens and sends them to a URL provided at runtime, which presents a risk of data exfiltration if directed to a malicious endpoint. Additionally, the deployment script dynamically loads and executes code from local directories.

  • Socket2mo

    2 alerts: gptAnomaly

  • Snyk2mo

    Risk: LOW · No issues

Signed by skilld at 820dfc8. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated last month
What it can do
Reads files Edits files Runs commands
version
2.30.0
author
Jeremy Longshore <jeremy@intentsolutions.io>
effort
high
All 6 allowed tools
ReadWriteEditGrepGlobBash(cmd:*)
Other metadata
argument-hint
<agent-name or project-id>
tags
[
  "ai",
  "deployment",
  "ci-cd"
]
compatibility
Designed for Claude Code

README badge

README badge for jeremylongshore/tons-of-skills-marketplace/adk-deployment-specialist