Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Adk Deployment Specialist · jeremylongshore | DecimalAI

jeremylongshore/adk-deployment-specialist

SkillScore68Maintained

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'.

agent-toolingby jeremylongshore

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

AI Engineer pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/jeremylongshore-adk-deployment-specialist/SKILL.md

decimalai skills pull jeremylongshore-adk-deployment-specialist --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/jeremylongshore-adk-deployment-specialist/SKILL.md https://app.decimal.ai/s/jeremylongshore-adk-deployment-specialist/SKILL.md

Raw file: /s/jeremylongshore-adk-deployment-specialist/SKILL.md ↗ · machine-readable summary at /s/jeremylongshore-adk-deployment-specialist.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 46d ago

Yes — +14% lift-7% tokens0% turns68% of whole cases pass with skill (15/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-04	✗→✓	▲ Improved	-9%	0%
case-17	✗→✓	▲ Improved	-56%	0%
case-19	✗→✓	▲ Improved	-2%	0%
case-06	✓→✓	＝ Same ✓	84%	0%
case-07	✓→✓	＝ Same ✓	-36%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/3/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

jeremylongshore / claude-code-plugins-plus-skills

skills/.curated/adk-deployment-specialist/SKILL.md

View on GitHub ↗

License MITImported 8/25/2026Version v1Bundled 8 files · scripts

Skill contents

scripts/

references/

SKILL.md

Adk Deployment Specialist

Overview

Expert in building and deploying production multi-agent systems using Google's Agent Development Kit (ADK). Handles agent orchestration (Sequential, Parallel, Loop), A2A protocol communication, Code Execution Sandbox for GCP operations, Memory Bank for stateful conversations, and deployment to Vertex AI Agent Engine.

Prerequisites

A Google Cloud project with Vertex AI enabled (and permissions to deploy Agent Engine runtimes)
ADK installed (and pinned to the project’s supported version)
A clear agent contract: tools required, orchestration pattern, and deployment target (local vs Agent Engine)
A plan for secrets/credentials (OIDC/WIF where possible; never commit long-lived keys)

Instructions

Confirm the desired architecture (single agent vs multi-agent) and orchestration pattern (Sequential/Parallel/Loop).
Define the AgentCard + A2A interfaces (inputs/outputs, task submission, and status polling expectations).
Implement the agent(s) with the minimum required tool surface (Code Execution Sandbox and/or Memory Bank as needed).
Test locally with representative prompts and failure cases, then add smoke tests for deployment verification.
Deploy to Vertex AI Agent Engine and validate the generated endpoints (/.well-known/agent-card, task send/status APIs).
Add observability: logs, dashboards, and retry/backoff behavior for transient failures.

Output

Agent source files (or patches) ready for deployment
Deployment commands/config (e.g., vertexai.Client.agent_engines.create() invocation + required parameters)
A verification checklist for Agent Engine endpoints (AgentCard + task APIs) and security posture

Error Handling

See ${CLAUDE_SKILL_DIR}/references/errors.md for comprehensive error handling.

Examples

See ${CLAUDE_SKILL_DIR}/references/examples.md for detailed examples.

Resources

ADK docs:
Workload Identity (CI/CD): https://cloud.google.com/iam/docs/workload-identity-federation
A2A / AgentCard patterns: see 000-docs/6767-a-SPEC-DR-STND-claude-code-plugins-standard.md

Illustrative Want to see what jeremylongshore/adk-deployment-specialist would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/13/2026, 01:24 UTC · 22 cases

Lift

+14%

68% vs 55%

Agent turns

1.0 vs 1.0

Tokens

-7%

2,215.73 vs 2,371.14

fail → pass · 3pass → pass · 12fail → fail · 7

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 I need to build a multi-agent workflow using Google's Agent Development Kit for automated data analysis and publish it to Vertex AI Agent Engine. Could you help structure the system, implement the agents, and deliver the ready-to-use agent source files, the Python deployment invocation script for Agent Engine, along with an endpoint and security verification checklist?	fail→fail	29,122	19,288	-34%	1	1	0%	6,211	4,805	-23%	0	0	—
case-02 We are converting our single customer support bot into a sequential ADK multi-agent setup where an intake agent passes tasks to a resolver using stateful memory. Please generate the updated agent source code, the commands needed to deploy to Vertex AI Agent Engine, and a post-deployment verification checklist covering the AgentCard endpoints and security posture.	fail→fail	15,500	12,282	-21%	1	1	0%	3,117	3,115	-0%	0	0	—
case-03 I want to deploy a cloud infrastructure management agent using ADK that leverages code execution sandboxes and communicates via A2A. Please guide the design and produce the finalized agent code patches, the Agent Engine client deployment config, and a testing checklist for task status polling and endpoint validation.	fail→fail	22,396	20,858	-7%	1	1	0%	4,983	4,961	-0%	0	0	—
case-04 We need to expose a standardized discovery card and inter-agent interface for our newly built orchestration nodes. Standard REST endpoints or OpenAPI definitions are often suggested, but we want the native pattern for cross-agent task submission and capability advertising. What endpoint path must be hosted to expose the capability metadata, and how are task status updates communicated?	fail→pass	14,767	10,390	-30%	1	1	0%	2,947	2,677	-9%	0	0	—
case-05 We want to execute three data extraction nodes simultaneously to query multiple databases, then aggregate their results into a summary report. Some engineers advocate managing this with custom Python threading loops, while others suggest using Vertex AI Reasoning Engine wrappers. What is the recommended runtime creation method for deploying this concurrent setup to Google's specialized agent runtime?	fail→fail	28,257	10,907	-61%	1	1	0%	2,933	2,699	-8%	0	0	—
case-06 We are designing a document drafting system where a writer agent creates content and a critic agent evaluates it, repeating the cycle until quality criteria are met. Which orchestration pattern should be configured for this iterative feedback workflow?	pass→pass	5,939	7,416	+25%	1	1	0%	1,009	1,858	+84%	0	0	—
case-07 Our multi-turn conversation agent needs to persist user preferences and past interaction context across separate user sessions without relying on ad-hoc SQL tables. Which native state persistence mechanism should be wired into the tool surface?	pass→pass	13,957	6,330	-55%	1	1	0%	2,504	1,593	-36%	0	0	—
case-08 Our operations agent needs to execute generated Python scripts to inspect cloud resources securely without risking direct system compromise on the host worker. What mechanism should be included in the tool surface for isolated execution?	pass→pass	11,410	6,573	-42%	1	1	0%	2,057	1,672	-19%	0	0	—
case-09 We are setting up a GitHub Actions pipeline to deploy agent runtimes to Google Cloud. A team member suggested generating a long-lived service account key JSON file and storing it as a GitHub Secret. What authentication mechanism should be used instead?	pass→pass	7,039	7,460	+6%	1	1	0%	1,469	2,042	+39%	0	0	—
case-10 Before creating deployment scripts for our multi-agent framework, what pre-deployment requirement must be checked regarding the local SDK installation relative to the target runtime?	pass→pass	6,461	3,224	-50%	1	1	0%	942	995	+6%	0	0	—
case-11 We have deployed our multi-agent application to the cloud runtime. Beyond confirming HTTP 200 on the main index, which specific task interface capabilities must be verified in the smoke test checklist?	pass→pass	13,242	8,308	-37%	1	1	0%	2,344	2,095	-11%	0	0	—
case-12 During peak load, underlying API calls from downstream tools occasionally hit HTTP 429 or transient 503 errors. What resilience mechanism and monitoring components must be included in the production deployment configuration?	pass→pass	14,231	14,115	-1%	1	1	0%	2,806	3,280	+17%	0	0	—
case-13 We have a three-step processing pipeline: Step A parses input data, Step B enriches the parsed record, and Step C formats the output into JSON. Which orchestration structure handles this linear chain?	pass→pass	4,070	4,657	+14%	1	1	0%	746	1,331	+78%	0	0	—
case-14 When defining tool access for a lightweight triage node that only routes incoming tickets based on metadata, should we attach full GCP administration tools and unrestricted bash runners?	pass→pass	8,914	6,848	-23%	1	1	0%	1,515	1,685	+11%	0	0	—
case-15 Before writing deployment code for a new multi-agent system, what three architectural elements must be explicitly defined in the agent contract plan?	fail→fail	7,819	4,428	-43%	1	1	0%	1,413	1,338	-5%	0	0	—
case-16 What testing step should be conducted immediately after local prompt verification and prior to issuing cloud deployment commands?	fail→fail	7,347	2,728	-63%	1	1	0%	1,181	974	-18%	0	0	—
case-17 What discovery URL path must be validated as part of the post-deployment endpoint verification checklist for standard inter-agent metadata?	fail→pass	10,471	2,271	-78%	1	1	0%	1,977	877	-56%	0	0	—
case-18 Our deployment script requires passing credentials to access external databases. How should sensitive credentials be managed to adhere to security posture standards?	pass→pass	12,681	11,458	-10%	1	1	0%	2,255	2,714	+20%	0	0	—
case-19 When beginning a new deployment specialist workflow, what is the initial step before writing any agent interface code or provisioning runtimes?	fail→pass	5,574	3,033	-46%	1	1	0%	999	977	-2%	0	0	—
case-20 How do I configure Google Cloud VPC Network Peering between a custom VPC network and a SQL database instance using Terraform?	fail→fail	12,814	8,901	-31%	1	1	0%	2,838	2,476	-13%	0	0	—
case-21 Our BigQuery SQL queries on partition-filtered log tables are running slowly. How should we optimize partition pruning and clustering keys in BigQuery?	pass→pass	17,849	10,289	-42%	1	1	0%	3,248	2,303	-29%	0	0	—
case-22 Can you help me write a multi-stage Dockerfile for a generic Go web server to minimize final container image size?	pass→pass	12,295	8,705	-29%	1	1	0%	2,671	2,279	-15%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +14 percentage points is the difference between those two pass rates over the 22 comparable cases.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 54.6%
Pass rate with the skill: 68.2%
Cases attempted: 22
Cases counted toward lift: 22
Cases that regressed: 0
Skill version measured: v1
Run completed: 8/13/2026, 01:24 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-04

Prompt

We need to expose a standardized discovery card and inter-agent interface for our newly built orchestration nodes. Standard REST endpoints or OpenAPI definitions are often suggested, but we want the native pattern for cross-agent task submission and capability advertising. What endpoint path must be hosted to expose the capability metadata, and how are task status updates communicated?

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run68

Live evalNeeds ≥10 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+1% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 46 days ago

Other measured skills in the registry, with their headline benchmark lift.