Canvas Adviser: Install, Source and Security | FunnelSlayer

Canvas Adviser

Published by newyngwieslash-ops in minimax-h3-codex-drama

No known issues1 installs

What this skill does

Set up, launch, troubleshoot, and use the local Codex Drama web canvas. Apply when the user asks about the Canvas plugin, its environment, vd-workflows, provider connections, project import/export, or Codex Plan text/image nodes. For full video production, use available Drama production skills.

Add Canvas Adviser to your agent

Review the source and files first. When you are ready, copy the prompt instruction or use the CLI command supported by your environment.

Install with a prompt

Paste this into a compatible coding agent:

add this skill "canvas-adviser" from https://github.com/newyngwieslash-ops/minimax-h3-codex-drama

Install with the CLI

Run this command in a controlled environment after reviewing the repository:

npx skills add https://github.com/newyngwieslash-ops/minimax-h3-codex-drama --skill canvas-adviser

Skill instructions

Canvas adviser

Help the user run and use the self-contained Canvas plugin. Resolve the plugin root as two directories above this SKILL.md's containing directory; it contains package.json, scripts/, and src/. Do not assume the current working directory or a fixed plugin-cache version.

For installation, launch, or environment failures, read environment.md. Run node <plugin-root>/scripts/doctor.mjs --json first; add --probe when checking existing services. Resolve missing prerequisites within the user's request, run the bundled setup helper when dependencies/build are absent, start the loopback server, verify /health, and open its URL in a Codex browser panel when available. Use an explicit stable data directory outside the plugin cache. Installation requests authorize the required npm dependency setup; do not repeatedly ask for permission already given.

For canvas operation, graph changes, workflow imports, or generation failures, read usage.md. Inspect the current project and provider/node catalogs before choosing IDs or editing it. Preserve explicit project and node model choices. New TEXT WORKFLOW and IMAGE WORKFLOW nodes use codex-plan by default, with the signed-in Codex catalog’s default model and the selected model’s default reasoning effort. Codex Plan uses the local Codex CLI sign-in and needs no Base URL or API key.

The browser adviser chat is a separate SDK conversation, not the current Codex app task. It advises on the graph and cannot directly mutate the live canvas. To change the canvas from this task, use the browser or the local JSON RPC helper. Edits are cached separately from the last explicit Save; projects/get may include a draft, and pending browser edits may be newer. Reconcile both before writing. New projects and example copies remain unsaved until Save; Discard restores the last save or removes a never-saved copy. Run generation when the user requests it; an environment check alone does not authorize a paid generation.

Distinguish a registered vd-node, its containing vd-workflow, and the executable comfyui-workflow selected by a ComfyUI provider. Importing a JSON node pack does not install Python custom nodes or model weights. Use available MiniMax H3 Drama skills for production planning, model/workflow setup, and finishing when the task needs them; Canvas setup and Codex Plan nodes do not require those companion plugins.

Report the usable URL, actual data directory, checks performed, and any specific remaining blocker. Never report generation capability as verified just because the model catalog is present. Do not read or print authentication files or API keys; codex login status is sufficient for sign-in diagnosis.

Files included

  • agents/openai.yaml
  • references/environment.md
  • references/usage.md
  • SKILL.md

More skills