Fabric CliSAFE

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

# Examples
"Sales.Workspace/Model.SemanticModel"
"Production.Workspace/LH.Lakehouse"
"Dev.Workspace/Report.Report"

python3 create_direct_lake_model.py "src.Workspace/LH.Lakehouse" "dest.Workspace/Model.SemanticModel" -t schema.table
python3 create_direct_lake_model.py "Sales.Workspace/SalesLH.Lakehouse" "Sales.Workspace/Sales Model.SemanticModel" -t gold.orders

python3 execute_dax.py "ws.Workspace/Model.SemanticModel" -q "EVALUATE VALUES('Date'[Year])"
python3 execute_dax.py "Sales.Workspace/Sales Model.SemanticModel" -q "EVALUATE TOPN(10, 'Orders')" --format csv
python3 execute_dax.py "ws.Workspace/Model.SemanticModel" -q "EVALUATE ROW(\"Total\", SUM('Sales'[Amount]))" -o results.json

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: fabric-cli
description: "Expert guidance for the Fabric CLI (`fab`) and the Fabric and Power BI REST APIs: workspaces, items, lakehouses, notebooks, pipelines, semantic models, reports, capacities, OneLake, deployment and admin. Also estimates the capacity units (CU) an operation will consume and its impact on the capacity before running it. Automatically invoke whenever the user mentions Fabric, Power BI Service, a Fabric or Power BI workspace, a capacity or F SKU, OneLake, `fab`, or asks to create, run, refresh, schedule, deploy or delete anything in Fabric, including preview items such as Plan, Ontology, Graph or Copilot."
---

# Fabric CLI

Guidance for using `fab` to programmatically manage Fabric & Power BI service

- Install via `uv tool install ms-fabric-cli` (get `uv` via `winget install uv` or `brew install uv`)
- Fabric CLI is for working with the Cloud environment and not local files; it works with Power BI Pro, PPU, or Fabric; you DO NOT need a Fabric SKU to use the Fabric CLI
- Keep `fab` current: check the installed version against the latest `ms-fabric-cli` release and upgrade with `uv tool upgrade ms-fabric-cli` unless the user has pinned a specific version. Discover commands and flags with `fab --help` and `fab <command> --help` rather than hard-coding behavior; the CLI surface changes regularly

> [!IMPORTANT] 
> Any time you encounter errors, user preferences or learnings when using the Fabric cli, ALWAYS note these down in the user memory rules, i.e. `.claude/rules/fabric-cli.md` for future improvement. 
> This is ONLY for generic learnings and not for item- or task-specific learnings.

## When to use this skill

- Use whenever the user mentions "Fabric" or "Power BI"
- Use when user asks about Power BI workspaces, deployment, tenants, publishing, download, permissions, or data


## Critical general rules

- IMPORTANT: The first time you use `fab` run check that it is up to date to the latest version (upgrade with `uv tool upgrade ms-fabric-cli` unless the user has pinned a version) and run `fab auth status`; If user isn't authenticated, ask them to run `fab auth login`
- Always use `fab --help` and `fab <command> --help` the first time you use a command to understand its syntax
- You must search the skill /references/ for relevant reference files that explain certain commands, examples, scripts, or workflows before you start using `fab`
- Before first use, ask the user if they have Fabric admin access, sensitivity labels or DLP policies, any API restrictions, or preferences for Fabric/Power BI API usage; remind user to add this to memory files
- If workspace or item name is unclear, ask the user first, then verify with `fab ls` or `fab exists` before proceeding
- Ensure that you avoid removing or moving items, workspaces, or definitions, or changing properties without explicit user direction
- IMPORTANT: Before any operation that consumes capacity units (creating items, running notebooks, pipelines or refreshes, large queries, Copilot, preview items such as Plan, Ontology or Graph, anything scheduled), estimate its CU-hours and its impact on the capacity, and get explicit user approval when the charge is flat or per session, exceeds a quarter of the capacity's daily CU-hours, recurs, cannot be estimated, or the capacity already throttles; a user naming a feature is not consent to its cost. See [Capacity cost](#capacity-cost-estimate-before-acting) and [capacity-cost.md](./references/capacity-cost.md)
- If a command is blocked in your permissions and you try to use it, stop and ask the user for clarification; never try to circumvent it
- Create output directories before export: `fab export` does not create intermediate directories; `mkdir -p` the output path first or the command fails with `[InvalidPath]`


### Capacity cost: estimate before acting

Bursting hides cost: operations finish fast and their CUs are smoothed into the next 5 to 64 minutes (interactive) or 24 hours (background). On a small F SKU one operation can use more than a day of capacity, and the carry forward then throttles every workspace on it until idle capacity pays it back, which on an F2 can take weeks.

- Check the capacity first: `fab ls .capacities -l` for SKU and state; the Monitoring hub capacity page for utilization, throttling and carry forward. Any carry forward means no headroom
- Estimate the operation in CU-hours and EUR, and as a share of the SKU's daily budget (an F<n> has 24 * n CU-hours a day; an F2 has 48)
- Know the flat charges: Plan (preview) bills 30-day sessions per user (Planner 847, Stakeholder 168, Viewer 37 CU-hours); creating and editing a plan by REST was billed as a Stakeholder session (168 CU-h), and a session cannot be ended or refunded
- Pausing (`fab stop`) bills the whole carry forward at once and clears throttling; scaling up burns it down faster at a similar total cost
- Full thresholds, formulas, known charges and where to see what an operation cost: [capacity-cost.md](./references/capacity-cost.md)

### Use `-f` (force) for non-interactive use

The `fab` CLI prompts for confirmation, so you **you must always append `-f`** to prevent this UNLESS sensitivity labels are enabled, in which case you must ask the user. Do this for the commands:

- `fab get -q "definition"` ; sensitivity label confirmation
- `fab export` ; sensitivity label confirmation
- `fab import` ; overwrite confirmation
- `fab cp` / `fab cp -r` ; overwrite and sensitivity label confirmation
- `fab rm` ; delete confirmation
- `fab assign` / `fab unassign` ; capacity/domain assignment confirmation
- `fab mv` ; rename/move confirmation


## Quickstart guide

You must read and understand the common list of operations with simple examples

0. Check the commands, syntax, and auth status: `fab --help` and `fab auth status`
1. Check if the item exists if the user gave the workspace and item name: `fab exists "spaceparts-dev.Workspace/spaceparts-otc-full.SemanticModel"`
2. Find an item by nam

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

declared (5 observation(s))

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (4)

raw = base64.b64decode(value, validate=True)

curl -s -X POST "https://login.microsoftonline.com/<tenant-id>/oauth2/v2.0/token" \

| All | Full access (view, edit, explore, reshare, copy) |

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh

Gates applied: no_behavioural_pass.