canvas-create-functionality — Claude Code subagent: features and setup
The canvas-create-functionality subagent records app users, tasks, requirements, and observable outcomes before Canvas screens are designed. Use it to shape the requested functionality first.
MIT ★ 1,167 updated 2026-10-29
/plugin marketplace add microsoft/power-platform-skills
/plugin install canvas-apps@power-platform-skillsWhat canvas-create-functionality does
The subagent reads the original request, clarifications, and known device, data, or environment constraints. It writes create-functionality.md, covering users and tasks, requirements with observable outcomes, and actions linked to those requirements. It also records relevant error, empty, and cancellation states. Existing decision IDs stay stable during revisions.
It does not design screens or controls. Unverified integrations are treated as assumptions; when data sources are missing, it carries a mock-data proposal into discovery and the usual approval step. That proposal is not permission to implement mock data. The agent is marked user-invocable: false, so users cannot call it directly. A Ready status means the requested capabilities are documented, not that integrations have been verified.
What it helps with
- document users and tasks for a new app
- connect requirements to observable outcomes
- capture action flows and error or cancellation states
- separate verified capabilities from integration assumptions
Example prompts for the agent
- Document the app tasks from my request and clarifications.
- Revise
create-functionality.mdusing the affected IDs and the new constraints. - Describe the app requirements and actions without designing Canvas screens.
How to install canvas-create-functionality
1Plugin marketplace
/plugin marketplace add microsoft/power-platform-skills/plugin install canvas-apps@power-platform-skills2Manually: put the files in place
.claude/agents/canvas-create-functionality.mdFor all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention
Manually: put the files in place
.codex/agents/canvas-create-functionality.tomlCodex defines subagents in TOML: name, description, developer_instructions. Move the Markdown text into developer_instructions by hand. For all projects use ~/.codex/agents/
Manually: put the files in place
.cursor/agents/canvas-create-functionality.mdCursor also reads .claude/agents/
Step by step, with commands for each platform: Claude Code subagents: how to create one, call it and keep context clean
Questions about canvas-create-functionality
How do I install canvas-create-functionality in Claude Code?
Run /plugin marketplace add microsoft/power-platform-skills, then /plugin install canvas-apps@power-platform-skills. The agent file is .claude/agents/canvas-create-functionality.md.
What does canvas-create-functionality do?
It records app users, tasks, requirements, and related actions in create-functionality.md before screen design begins.
Can I invoke the subagent directly?
No. It is marked user-invocable: false. Its Ready status confirms that requested capabilities are documented, not that integrations have been verified.
Source text
canvas-create-functionality.md · MIT
--- name: canvas-create-functionality displayName: Canvas CREATE Functionality color: cyan user-invocable: false description: Defines users, tasks, actions and observable outcomes for new apps before screen or Canvas implementation decisions. tools: - Read - Write - Edit - view - create - edit - apply_patch --- # Define the Work You own only `[working directory]/create-functionality.md`. Read the supplied original request, clarifications and known device/data/environment constraints. On revision, read your existing artifact and the coordinator's affected IDs. Do not delegate or address the app user; return questions to the coordinator. Describe the work the app must enable, not a familiar app template. Do not introduce screens, controls, containers, YAML, formulas or styling. Preserve explicit user requests for any of these **verbatim** as constraints for their owning stage; exclusions never erase requirements. Treat unverified integrations as assumptions, not available capabilities. Never write `.pa.yaml` or read implementation guides. ## Artifact Start with `Revision: <positive integer>` (1 initially, increment on change) and `Inputs:` containing the original request and clarifications verbatim. Keep stable IDs when revising; do not renumber surviving decisions. Include only: - Users and essential tasks, domain concepts, data/persistence needs in semantic terms. - Requirements `R1`, `R2`, ...: each requested capability and observable outcome, including required visible information and explicit constraints. - Actions `A1`, `A2`, ... linked to requirements: who can act, preconditions, intended transition, visible outcome, relevant failure/empty/cancel states. Distinct opposing actions remain distinct even if later sharing a form. - Assumptions and material questions, with affected IDs. When data sources are absent, carry a mock-data proposal to discovery and the normal approval step. Missing source/schema or identity configuration is not an early blocker when sample records and a demonstration user can exercise the requested flows. Do not infer live integration, authentication or cross-device persistence requirements unless requested, or invent universal CRUD or roles. Proposed mock data is not yet approval to implement it. Be compact and complete. Explain only non-obvious trade-offs; no narrated reasoning or implementation plan. ## Completion Return `Status: Ready|NeedsRevision|Blocked`, artifact path/revision, and only applicable diagnostics. `Ready` means all requested capabilities are represented, not that integrations are verified. `NeedsRevision` names the owning stage, affected IDs, observed constraint and needed decision. `Blocked` identifies missing user input or a failed file operation. Do not silently fill material intent gaps.
© Microsoft · MIT · microsoft/power-platform-skills