Flows Skills

Agent-agnostic AI skills for Flows apps. Works with Claude Code, Cursor, Copilot, and any agent that supports the skills format.

Install

Pull all skills into your Flows app:

npx @cognite/cli@latest apps skills pull

Pull a specific skill:

npx @cognite/cli@latest apps skills pull --skill integrate-fusion-agent

Certification flow

Four skills drive the Flows app certification flow end to end. Run them in order:

1. flows-app-brief — Right after npx @cognite/cli apps create. Coaches you through the App Brief and writes App-Brief.md.
2. flows-code-review — Technical review. Writes reviews/code-review/feedback-round-<N>/. Re-run until 0 Must Fix.
3. flows-design-review — Manual design quality assessment (10 questions). Writes reviews/design-review/feedback-round-<N>/. Target average ≥ 3.8.
4. flows-external-app-submit — Verifies the prior three artifacts and runs npx @cognite/cli apps submit.

Available Skills

Skill	Description
flows-app-brief	Certification coach (step 1) — captures app details, persona, problem, and success criteria, writes App-Brief.md
flows-review-checks	Shared technical-review bar (hunts, coverage scope, public criteria). Loaded by flows-code-review and by external review skills — do not copy
flows-code-review	Local technical review runner (step 3) — loads flows-review-checks, writes reviews/code-review/feedback-round-<N>/
flows-design-review	Manual design quality assessment (step 4) — scores the 10 quality-guidelines questions, writes reviews/design-review/feedback-round-<N>/design-review-report.md
flows-external-app-submit	Submission gate (step 5) — verifies brief + code review (0 Must Fix) + design review (avg ≥ 3.8), then runs npx @cognite/cli apps submit
integrate-fusion-agent	Default AI path — Atlas / EOS sidebar
integrate-atlas-chat	Exception — in-app useAtlasChat when the EOS sidebar cannot work
create-client-tool	AtlasTool for an approved in-app chat (sidebar tools: integrate-fusion-agent)
setup-python-tools	Pyodide for in-app chat only
call-cognite-function	Invokes a deployed CDF Function by ID or external ID — session nonce, call, poll until complete, return typed result
code-quality	Reviews Flows apps for code quality, maintainability, and clean code issues
correctness-and-error-handling	Reviews for bugs, missing error states, unhandled rejections, and edge cases
dm-graph-traversal	CDF Data Modeling graph traversal patterns — query-vs-list decisions, traversal payload guardrails, failure debugging, and payload-shape test requirements
dm-limits-and-best-practices	DMS concurrency, pagination, batching; LLM-over-rows cap (5 / max 50)
integrate-file-viewer	Integrates CogniteFileViewer to preview CDF files (PDFs, images, text)
performance	Optimizes Flows apps for speed, render counts, and bundle size
pull-changes-resolve-conflicts	Merge or rebase workflow — list conflicts, analyze ours/theirs, get user approval before resolving
security	Reviews for security issues — credentials, user input, external data
migrate-app-to-flows	Migrates a legacy Dune POC app to Flows Custom App infrastructure — fixes app.json, auth, manifest.json, and deploy scripts
setup-flows-auth	Migrates React apps to Flows auth or adds DuneAuthProvider
hide-show-shell	Hides/reveals the Fusion sidebar + topbar for full-screen app-only mode via api.setHideShell(), with best practices for always giving users a way back
handle-startup-state	Reads initialState from connectToHostApp() on mount — deep links, shareable URLs, and host-provided startup arguments — with safe parsing, defaults merging, and optional syncInternalState write-back
design	Aura UI — components and tokens, layouts, UX copy, forms/async feedback, accessibility (skills/design/)

Contributing

Add a new skill by creating a skills/<name>/SKILL.md file with proper frontmatter:

---
name: my-skill
description: "When to use this skill..."
allowed-tools: Read, Glob, Grep, Edit, Write
---

Consolidated Aura guidance uses the design skill (skills/design/SKILL.md). Older per-topic design-* skills were merged into that folder.

Authoring style: Use ## Section markdown headings for structural sections like Role, When to use, and Setup. XML tags (e.g., <rubric>, <example>) are acceptable where they serve a machine-parsing purpose.