claude-power-automate

A Claude Code suite for Microsoft Power Automate development — eight focused Agent Skills plus a local, open CLI for inspecting and deploying cloud flows. No third-party hosted service, no stored secrets.

⚠️ Beta (v0.3)

This is early, pre-1.0 software. The CLI surface, command output, skill contents, and config schema may change without notice between 0.x releases. Pin to a tag if you depend on it.

Not affiliated with or endorsed by Microsoft or Anthropic. "Power Automate", "Power Platform", and "Dataverse" are trademarks of Microsoft; "Claude" is a trademark of Anthropic.

What's in the box

This suite gives an AI coding agent (Claude Code) — and you, at the terminal — everything needed to build, diagnose, and ship Power Automate cloud flows from local artifacts, the same way teams already manage code.

• power-automate CLI — a local Python client backed by the Dataverse Web API and the Power Platform API. Pull a live flow to a JSON artifact, diff it, back it up, deploy it (with verify), create new solution-aware flows, list environments/flows, and read run history for forensics.
• Eight Claude Code Agent Skills — domain knowledge and guardrails that make Claude reliable on Power Automate work instead of guessing.

The skills

Why

• Local and open. The runtime is plain Python plus official Microsoft APIs. No subscription and no session with a third-party hosted service are required to deploy.
• No secrets, ever. Short-lived OAuth tokens come exclusively from your existing az login session. The tool stores nothing. The only thing it reads is the non-secret target of a command (which environment / flow).
• Versioned flows. Treat definition + connectionReferences as artifacts you can diff, review, back up, and deploy through a DEV → PROD pipeline.

Install

CLI

Requires Python 3.10+ and the Azure CLI.

pip install "git+https://github.com/Wals-pro/[email protected]"
power-automate --help

Or from a clone:

git clone https://github.com/Wals-pro/claude-power-automate
cd claude-power-automate
pip install -e .

Skills (Claude Code)

Copy (or symlink) the skills you want into your Claude Code skills directory, e.g. your project's .claude/skills/:

cp -R skills/power-automate-* /path/to/your/project/.claude/skills/

Claude will load each SKILL.md and pull in its references/ on demand.

Quickstart

# 1. Authenticate (once) with an account that can manage the environment.
az login            # see docs/az-cli-setup.md for cross-tenant / MFA notes

# 2. Point the CLI at a target (flags, env vars, or a profile — see below).
export POWER_AUTOMATE_ENVIRONMENT_ID="00000000-0000-0000-0000-000000000000"
export POWER_AUTOMATE_FLOW_ID="11111111-1111-1111-1111-111111111111"

# 3. Inspect, then pull the live flow to a local artifact.
power-automate status --profile acme
power-automate pull   --profile acme --output flow.json

# 4. Edit flow.json, preview the change, deploy safely (backup + verify).
power-automate diff   --profile acme flow.json --unified
power-automate deploy --profile acme flow.json --dry-run
power-automate deploy --profile acme flow.json

deploy always reads the live flow first, writes a timestamped backup under .power-automate-backups/<profile>/, patches only the flow definition, and then reads the flow back to verify the result.

Configuring targets

The CLI never stores secrets. It only resolves the non-secret target of a command — environment_id, flow_id, dataverse_url, entra_tenant_id — in this precedence order:

1. CLI flags: --environment-id, --flow-id, --dataverse-url, --entra-tenant-id
2. Environment variables: POWER_AUTOMATE_ENVIRONMENT_ID, POWER_AUTOMATE_FLOW_ID, POWER_AUTOMATE_DATAVERSE_URL, POWER_AUTOMATE_ENTRA_TENANT_ID (also POWER_AUTOMATE_<PROFILE>_<FIELD>)
3. profiles.json at ~/.config/power-automate-cli/profiles.json (override with $POWER_AUTOMATE_CONFIG), selected with --profile:

{
  "acme": {
    "environment_id": "00000000-0000-0000-0000-000000000000",
    "flow_id": "11111111-1111-1111-1111-111111111111",
    "dataverse_url": "https://org.crm4.dynamics.com",
    "entra_tenant_id": "22222222-2222-2222-2222-222222222222"
  }
}

dataverse_url is optional — if omitted, the CLI discovers it from the Power Platform environment metadata. entra_tenant_id is only needed when the environment lives in a different Entra tenant than your default az login.

Embedding

Host applications can inject their own (e.g. secrets-manager-backed) lookup without this package depending on it:

from power_automate_cli.config import register_profile_resolver

register_profile_resolver(lambda profile, field: my_store.get(profile, field))

Command reference

Run power-automate <command> --help for the full flag set.

Solutions & environment variables

The suite is built for solution-based ALM. List solutions and inspect their components, then read and set the environment variables that parameterize a solution across DEV/PROD:

power-automate solutions --profile acme
power-automate solution-components --profile acme --solution acme_CoreAutomation
power-automate env-vars --profile acme --solution acme_CoreAutomation
power-automate env-var-set --profile acme acme_ApiBaseUrl "https://api.example.com" --solution acme_CoreAutomation --dry-run

Secrets stay secret. Secret-type environment variables are Azure Key Vault-backed; the CLI masks their values by default (--reveal-secret shows only the Key Vault reference, never the secret), and env-var-set refuses Secret-type variables so a secret never reaches your shell history or Dataverse as plaintext.

See docs/az-cli-setup.md for authentication and docs/operations.md for the deploy runbook.

Development

pip install -e .
python -m unittest discover -s tests -p "test_*.py" -v

The test suite mocks all network and Azure CLI calls — no live tenant required.

License

MIT © Wals.pro. See CHANGELOG.md for release notes.