tool-selection
/tool-selection
Pick the right tool for each Power Platform task before acting - purpose-built plugin tools and MCP servers first, then the Dataverse SDK, Web API or pac CLI, and browser automation (Playwright) only for what has no API - and never tell the user something can't be done while browser automation could do it. Prevents the two failure modes: driving the maker portal through a browser for work an SDK does in one call (slow, brittle, costly), and refusing work that only the UI can do. Trigger whenever several tools could do the job: "use Playwright", "open the browser", "click through the portal", "
- Author
- powercademy ↗
- Repository
- powercademy-skills ↗
- Licence
- MIT
- Last repository change
- 12.11.2026
Assessment and security
- Files in this package
- 1
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 12.11.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
What it does and when to use it
Pick the right tool for each Power Platform task before acting - purpose-built plugin tools and MCP servers first, then the Dataverse SDK, Web API or pac CLI, and browser automation (Playwright) only for what has no API - and never tell the user something can't be done while browser automation could do it. Prevents the two failure modes: driving the maker portal through a browser for work an SDK does in one call (slow, brittle, costly), and refusing work that only the UI can do. Trigger whenever several tools could do the job: "use Playwright", "open the browser", "click through the portal", "
Triggers from the description
Original author description
Pick the right tool for each Power Platform task before acting - purpose-built plugin tools and MCP servers first, then the Dataverse SDK, Web API or pac CLI, and browser automation (Playwright) only for what has no API - and never tell the user something can't be done while browser automation could do it. Prevents the two failure modes: driving the maker portal through a browser for work an SDK does in one call (slow, brittle, costly), and refusing work that only the UI can do. Trigger whenever several tools could do the job: "use Playwright", "open the browser", "click through the portal", "
How it works
- The ladder
- Why the order matters
- Common pairings
- Anti-patterns
- Say which tool, and why
- Grow
SKILL.md preview
---
name: tool-selection
description: >
Pick the right tool for each Power Platform task before acting - purpose-built plugin tools and MCP servers first, then the Dataverse SDK, Web API or pac CLI, and browser automation (Playwright) only for what has no API - and never tell the user something can't be done while browser automation could do it. Prevents the two failure modes: driving the maker portal through a browser for work an SDK does in one call (slow, brittle, costly), and refusing work that only the UI can do. Trigger whenever several tools could do the job: "use Playwright", "open the browser", "click through the portal", "is there an API for this", "bulk update records", "load this data", "create these columns", "change this setting", "check it looks right", "why is this so slow", "can you do this", and before saying any Power Platform task is impossible. Also trigger when browser automation is about to be used for data, schema, solution or flow work. When in doubt, trigger.
---
# Tool Selection — the right tool for the job
An agent with a browser can do almost anything on the Power Platform — slowly.
An agent with only APIs is fast — until it hits a setting with no API and
declares the task impossible. Both are wrong. The job is to reach for the most
direct tool that can do the work, and to keep going down the ladder rather
than give up.
## The ladder
Work top-down. Stop at the first rung that can do the job.
1. **Purpose-built tools for the workload** — the Microsoft plugin skills and
MCP servers built for it (flows, Dataverse, canvas apps, Power Pages,
Copilot Studio). They know the platform's rules and validate as they go.
2. **SDK, Web API or CLI** — the Dataverse SDK or Web API for data and schema
(batched for anything bulk), the pac CLI for solutions, environments, auth
and components, the Azure CLI for Azure. Seconds, repeatable, scriptable,
auditable.
3. **Browser automation** — for what has no API: UI-only settings, admin-centre
toggles, maker-portal-only features, and anything where *the UI itself* is
the thing being checked (does the form render, does the site look right,
what does the user actually see).
Concrete tool names per rung live in `${PLUGIN_ROOT}/shared/microsoft-refs.md`
under "Tool map".
**Browser automation is the last resort, not never.** If rungs 1 and 2 can't do
it, use rung 3 — don't tell the user it can't be done. If no browser
automation is available either, give them the exact click path instead.
## Why the order matters
- **Speed.** One SDK call creates a column; the browser route is a dozen page
loads, waits and clicks.
- **Reliability.** Pages change, load slowly, throw pop-ups. A script that
worked yesterday fails today on a renamed button. APIs don't move like that.
- **Cost.** Every browser step returns a snapshot of the page into the
conversation. Twenty steps is twenty page dumps.
- **Repeatability.** An SDK or CLI script can be re-run, reviewed and put in a
pipeline. A browser session can't be promoted to test and production.
- **Auditability.** API calls leave a clear record of what changed; clicks
leave a recording at best.
## Common pairings
| Task | Reach for | Fall back to |
|---|---|---|
| Create or change tables, columns, choices, relationships | Dataverse tools / SDK / Web API | Maker portal in the browser |
| Load, update or delete many records | SDK or Web API, **batched** | Never record-by-record in a browser |
| Query or read data | Dataverse tools / SDK | — |
| Export, import or promote a solution; environment operations | pac CLI or native pipelines | Admin centre in the browser |
| Build, edit or debug a cloud flow | Microsoft's flow tooling | Designer in the browser |
| Edit a canvas app | Microsoft's canvas authoring tooling | Studio in the browser |
| Check a deployed site, form or app *looks and behaves right* | **Browser automation** — the right tool here | Ask the user for a screenshot |
| A tenant or environment setting with no API | Browser automation | Exact click path for the user |
| Verify a fact about the platform | Microsoft Learn | — |
## Anti-patterns
- Opening the maker portal to create a column the SDK could create in one call.
- Clicking through records one by one to update them.
- Taking screenshots to *read data* that can be queried.
- Using the browser to "double-check" what an API call already confirmed —
unless the point is to check the UI itself.
- Saying "I can't change that setting" when browser automation is available.
- Changing environment, security or tenant settings through the browser without
explicit consent — UI-only changes are still changes, and harder to undo.
## Say which tool, and why
Before acting, one line: which rung, and why — especially when choosing the
browser. *"There's no API for this setting, so I'll use the browser — slower,
but it works."* When the browser is the right call for verification, say that
too: *"Changes made through the Web API; opening the form in the browser to
confirm it renders."* API to change, browser to see — that pairing is often
the best of both.
## Grow
When a task teaches a better pairing — a new API that replaces a browser
route, or a UI-only setting worth knowing about — offer to add it to the table
above, the same way the other conventions grow.
Versions
| Repository | Score | Licence |
|---|---|---|
| This page · powercademy/powercademy-skills | 90 | MIT |
| Open-RUA/uni-robot-agent | 100 | MIT |
Database snapshot: 13.11.2026 · package version b5d434916194 (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 100dataverse-overview
Background orientation for Microsoft Dataverse work: the dataverse skill map, hard rules, tool-capability matrix (MCP, Python SDK, dataverse CLI, PAC CLI, Azure CLI, GitHub CLI, Web API), SDK method cheat-sheet, and the safe change lifecycle. Load first for any Dataverse task,… - 100power-apps-code-apps
Use when building, scaffolding, debugging, or deploying Microsoft Power Apps Code Apps using React, Vue, or TypeScript in VS Code. Handles project initialization via pac CLI, Dataverse and connector data access via the @microsoft/power-apps SDK, power.config.json configuration,… - 100power-platform
Builds, ships and verifies Microsoft Power Platform work the way software is built: canvas apps as .pa.yaml in git, Dataverse tables, choices, roles and solutions, Power Automate cloud flows, pac pack and import, reporting from app history, and Power Pages sites (pages in git,… - 90code-apps
Use for Power Apps Code Apps - the pro-code model where a standalone Vite-built React/Vue/TS single-page app uses the `@microsoft/power-apps` client library and is pushed into a Power Platform environment with no server to manage. Covers scaffolding from the official Vite… - 100playwright-cli
Automate browser interactions, test web pages and work with Playwright tests using the playwright-cli command. Use when the user wants to open a browser, navigate pages, click elements, fill forms, take screenshots, run Playwright tests, debug browser automation, or perform any… - 100ac-qa-playwright-cli
Browser automation via playwright-cli. Token-efficient alternative to Playwright MCP - uses CLI commands through Bash instead of MCP tool schemas. Supports navigation, interaction, screenshots, video recording, and session management. Triggers on keywords: browser, e2e,…