skills

Authoring standalone HTML artifacts — `scripts/<name>/main.html` documents that Fused and Flow serve full-bleed and wire to a `window.fused` runtime (callUdf / runSql / params). Use when the output of a project is a bespoke interactive HTML page — a custom chart, a form, a report, a small app — that calls UDFs and runs SQL directly, rather than a JSON-UI widget. Covers the `window.fused` API, the `{{ref}}`/`$param` data grammar, load-order gotchas, host differences, and how the artifact gets rendered in front of a human.

Show the human a real browser UI — to ask a question, get an approval/decision, or review a plan — built from Fused's JSON-UI primitives and opened with `fused widget open` (one-shot — inline `--config` or a `.json` file) or the parley (`widget push`/`widget watch`, standing). Use in Claude Code whenever a structured choice, form, approval, or plan review would be clearer than plain terminal text, and you want the human's answer back as JSON. If the task touches an existing project, UDF, or data-bound widget (anything with `{{ref}}`/`sql`), also load `fused-projects` and `fused-widgets` first.

Reference for the fused CLI — environment management, file storage, secrets, code execution, and infrastructure commands. Use when writing or explaining shell commands that invoke `fused`, or when helping users set up, switch between, or provision environments. If the commands are part of building or running a project, load `fused-projects` first for the end-to-end model.

Authoring and previewing JSON-UI widgets as the response of running a project — the py-UDF-computes → json-widget-visualizes pattern, the {{ref}}/$param data grammar, how resolution runs through the compute backend, and the CLI surfaces (widget open, parley, deployed URL) that put a rendered widget in front of a human. Use whenever the desired output of a UDF/project is a widget, not raw data.

Best practices for running code through fused's execute_code tool. Use when writing or reviewing any mcp__openfused__execute_code call — covers how to structure user code, choose a data library, handle results, and write outputs to the file store. For security scanning, spec checks, and testing see fused-verify. If this is part of building or running a project, load `fused-projects` first for the end-to-end model (and `fused-widgets` if the result is a rendered UI).

The fused storage and secrets MCP tools — inspecting cloud-native datasets and managing secrets. Use when finding/listing/counting S3 objects, reading a Parquet/Arrow/CSV schema, minting a download URL, uploading content, or storing/reading/deleting secrets, via mcp__openfused__{list_files,count_files,get_file,get_file_schema,upload_file,get_secret,put_secret,list_secrets,delete_secret}. For running code over the data see fused-execute; for the equivalent CLI commands see fused-cli. If this is a step in building or running a project, load `fused-projects` first for the end-to-end model.

Security scanning, testing, and correctness validation for code running in openfused. Use when writing or reviewing verify_code, test_code, get_audit_log calls, or when advising on security policy, spec checks, data expectations, or code quality in the fused context. If you're validating code that belongs to a project, load `fused-projects` first for the end-to-end flow.

Reference for the infrastructure managed by fused — what resources exist, why each one is needed, and when they are created, updated, or deleted. Covers the AWS backend (IAM, Lambda, ECR, S3) and the local backend (data directories + venvs, see "Local backend infra"). Use when helping users understand, provision, or troubleshoot the resources that back an fused environment.

The canonical end-to-end guide for an agent driving Fused — pick an environment, create a project, decompose a task into UDFs, author specs and code, validate + commit, run/preview locally (often as a rendered widget), and deploy through preview to release. Code is authored by the driving agent (no codegen command, no API key); fused supplies validation, the spec↔code pairing hook, and run/deploy. Use to take a user request from prompt to a running, viewable result.

Step-by-step guide for installing and setting up fused for the first time, including AWS credential checks, detecting an existing installation, provisioning infrastructure, and verifying the setup. Use when a user asks how to install, configure, or get started with openfused.

JSON schemas for Fused canvas UI widget components (text, text-input, button, dropdown, charts, maps, sql-table, form, transformer, etc.). Use when authoring or editing widget JSON files (e.g. `widget_*.json` in canvas directories), validating widget props, or answering questions about which fields a given widget type accepts.

Entry-point router for the Fused skills in this repo. Use when a user asks how to get started with, set up, or install Fused, how to build a project or UDF, how to get a widget UI / dashboard out of Fused, or is unsure which Fused skill to load. Maps a goal to the specific skill(s) to read next.

Use when a Fused canvas.toml contains (or needs) a [[comments]] block — e.g. the user asks to check/fix/resolve canvas comments, review feedback pins on a canvas, or leave comments on a canvas for someone else. Covers reading, writing, and resolving comments via the fused CLI pull/push workflow.

Format reference for `canvas.toml` files that define a Fused canvas (nodes, edges, viewport) plus the surrounding folder layout (`{udfName}.py`, `.json`, `.md`, `.html`, `_shared.fused`). Use when creating, editing, or validating any `canvas.toml`, adding/removing UDF nodes, wiring edges, defining folder groupings, or scaffolding a new canvas folder.

Reference for the legacy Fused Python SDK command line interface, now namespaced under `fused workbench`. Use when the user asks how to run, push, share, or otherwise manage UDFs, canvases, files, or secrets via the Fused workbench CLI, or when authoring shell commands that invoke `fused workbench`.

Reference for using Fused's built-in integration connections inside UDFs. Covers data sources (Snowflake, BigQuery, GCS, S3, Airtable, Notion, Google Drive), compute/inference providers (Modal, Hugging Face, Baseten, Daytona, ComfyOrg, Slack), and LLM providers (Anthropic, OpenAI) — the fused.api connect helpers, secrets access, and common operations (query, write, list, invoke, infer). Use when the user is writing a UDF that reads from, writes to, or calls out to a connected service.

Orientation to what Fused is and when to use it. Use when deciding whether to use Fused for a task, planning a new Fused project, or understanding Fused's capabilities as a remote Python execution platform.

Comprehensive guide for writing Fused User Defined Functions (UDFs). Use when helping users create, optimize, or debug UDFs, including function structure, parameter handling, performance optimization, agent-friendly design, and security best practices.