Choose a useful implementation scope
Inspect three working examples before choosing an AI or search engagement. Each includes the input, recorded result, source code and a repeatable command. Take the one-page qualifications sheet to your team.
01 / MCP integration
Give an assistant a workflow it can inspect
A tool-connected assistant needs a clear contract between its client and the system it reads. This example connects a small Node client to the real Suede Creator Skills MCP server, discovers its tools and retrieves a named workflow.
5 protocol requests; 9 discoverable tools
The client completes discovery, search and workflow retrieval, then checks that an unknown skill returns an error.
Input and observed result
The client negotiates protocol 2025-06-18, routes a technical SEO review request and retrieves the named workflow instructions. The runner verifies the server, catalog and workflow source against commit 46445415a767. All discovered tools carry read-only annotations.
The saved report contains the exact search text, returned workflow names and instruction hash so another run can check the same response.
Training connected to the work
Claude Academy / Anthropic: Claude API, Claude Code, MCP, Amazon Bedrock and Google Cloud. Microsoft Learn: Azure AI Agents, Foundry IQ, Visual Studio Code, Microsoft 365 and Agent Framework. Hugging Face: Fundamentals of Agents. Inspect the issued training records.
02 / Technical SEO walkthrough
Find a source-page defect before it reaches a buyer
A useful source review connects a specific observation to a fix someone can verify. This walkthrough checks the actual Suede AI evidence page, then repeats the check on a controlled copy with three changes.
8 of 8 checks passed
Evidence page checked 2026-10-21. The committed HTML was checked locally; no HTTP request was made.
- Document title: pass
- Description: pass
- Canonical address: pass
- Indexable directive: pass
- One main heading: pass
- Parseable structured data: pass
- Responsive viewport: pass
- Reader next step: pass
Reproduce the repair check
The controlled copy removes the title and canonical address and adds a noindex directive. The same checker detects all three changes. Restoring the original source returns all eight checks to pass. The report retains the page address, capture method, date and SHA-256.
These checks describe the source page. Search impressions, answer citations and inquiries have their own measurement records, which can be compared after source repairs are published.
Training connected to the work
Semrush Academy: AI search, keyword research, on-page SEO, local SEO and digital PR. HubSpot Academy: AEO Fundamentals certification. Inspect the issued training records.
03 / Answer evaluation
Check the facts an AI answer must preserve
Define the acceptance criteria before choosing an output. This example uses the published credential inventory to check the holder, separate award totals, issuer, award type, source link and relationship in a structured answer.
4 of 4 expected decisions matched
Four hand-authored answer fixtures were evaluated against eight checks each.
- Source-aligned answer: 8/8 checks; accept.
- Conflated course and training totals: 5/8 checks; review.
- Missing verification link: 7/8 checks; review.
- Training rewritten as partnership: 7/8 checks; review.
Use the rubric with your own model
The source-aligned fixture is accepted. Other fixtures change the course and training totals, remove the verification link or rewrite personal training as a company partnership. The checker identifies those failures. When evaluating a model response, retain its full prompt, model, settings, date and original output alongside the score.
Source-grounded prompt · Authored fixtures · Evaluation code
Training connected to the work
IBM SkillsBuild: Craft Precise Prompts for AI Models. Google Skills: introductory generative AI and large language models. Databricks Academy: Generative AI Fundamentals accreditation. Inspect the issued training records.
Run all three demonstrations
Use Node 22 or newer and the pinned public skill-pack source. The recorded run used zero paid model calls.
git clone https://github.com/JasonColapietro/suede-creator-skills.git ../suede-creator-skills-demo
git -C ../suede-creator-skills-demo checkout --detach 46445415a767df5827f5e124f41749ef369cca83
node examples/training-proof/run.mjs --skills-repo ../suede-creator-skills-demo --out /tmp/suede-training-proofRun from a checkout of the example repository. Add --live to compare the current evidence page with the checked-out source.
Download the dated result · Download the qualifications sheet