AI WORKFLOW · GOOGLE WORKSPACE
From “Chatting” to“Getting Things Done”
A practical guide to the concept, structure, and execution of an AI Skill—using feature-release to show how it works.
▾ feature-release/
├── SKILL.md
├── scripts/
│ └── run-tests.js
└── templates/
└── test-report.md
A Skill is an SOP for AI
Large language models are excellent at understanding and generating content. What organizations need, however, is AI that can complete an entire task consistently and repeatably under defined rules.
SKILL ANATOMY
The three building blocks of a Skill
A Skill specifies not only what the AI should do, but how to do it, which tools to use, and when it must stop.
SKILL.md
Defines triggers, execution order, success criteria, and non-negotiable boundaries.
Example: When the “Feature Release” Skill is triggered: 1. Run npm test; 2. Record errors and the pass rate in test-report.md; 3. If tests miss the threshold, block the release and suggest fixes.
/feature-release/SKILL.mdscripts
Run tests, collect results, and generate reports with less uncertainty than ad hoc commands.
/scripts/run-tests.jstemplates
Standardize reports, approval records, and checklists so every result is consistent and auditable.
/templates/test-report.mdFEATURE-RELEASE · IN ACTION
One request triggers a complete release loop
- 01Match the SkillRecognize intent and load feature-releaseMATCH
- 02Run testsCollect pass rates, errors, and failed casesTEST
- 03Generate reportWrite the results to test-report.mdREPORT
- 04Apply the gateStop below threshold and recommend fixesGUARD
- 05Approve releaseContinue only after tests and approvals passRELEASE
GOOGLE WORKSPACE
Tools provide capabilities;
Skills define the workflow
A Skill does not automatically receive cloud access. Only authorized tools allow the AI to read, update, and organize Workspace content.
Docs
Read meeting notes and generate reports from templates.
Sheets
Aggregate data, calculate metrics, and build charts.
Drive
Locate, name, organize, and archive files.
Slides
Turn reports into executive presentations.
Gmail
Generate an email draft from the results and send it after confirmation.
BEFORE · AFTER
What changes with GWS integration?
| WORK AREA | WITHOUT GWS INTEGRATION | WITH AUTHORIZED GWS INTEGRATION |
|---|---|---|
| Docs | Copy text into AI, then paste the result back manually | Read a specified document and suggest or write changes within permissions |
| Sheets | Export or upload data for offline analysis | Read sheets, update cells, and generate summaries |
| Drive | Find, download, upload, and archive files manually | Search, classify, and create files within authorized scope |
| Gmail | Generate email copy for the user to paste and send | The user can ask AI to create an email draft, then send it to a specified email address |
| Cross-app work | People move information between every step | A Skill connects apps and hands data between them |
| Risk control | Users perform most checks manually | Permissions, approval gates, and activity records support governance |
FROM ANSWERS TO OUTCOMES
Turn working methods into executable capabilities
A Skill defines what to do, how to do it, and what to produce. Authorized tools perform the cloud actions. Together, they move AI from content assistant to reliable collaborator.