> For the complete documentation index, see [llms.txt](https://docs.ui-color-palette.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.ui-color-palette.com/claude/agents.md).

# Agents

The plugin ships six agents. `uicper` is the default orchestrator — it delegates to the five specialized agents when a task falls within their scope.

***

## uicper

**Default agent** — configured in `settings.json`.\
**Model**: Sonnet · **Effort**: high · **Max turns**: 30

Top-level orchestrator for multi-step workflows. It coordinates skills, delegates to specialized agents, and assembles results across the full 4-phase workflow.

**Typical workflows it handles**:

* generate → audit → export
* generate → preview → sync to design tool
* retrieve from community → rebuild → audit → export

It does not perform heavy computation itself — it picks the right workflow and delegates.

***

## palette-color-systemer

**Model**: Sonnet · **Effort**: medium · **Max turns**: 20

Specialized agent for guided semantic color system design. Walks the user through the full taxonomy → binding → submission workflow step by step.

**Responsibilities**:

* Suggest taxonomy patterns (Role × State, Role × Prominence × State, Surface × Content, Custom)
* Confirm members per group before proceeding
* Propose intelligent default bindings derived from palette color ids and shade stops
* Flag tokens that have no natural binding in the current palette
* Batch all adjustments before calling `get_color_system` once
* Display the resolved token matrix and route to the next deploy step

**Hard prerequisite**: `PaletteData` or `base` + `themes` must be in context. If missing, the agent stops and redirects to `scale-palette` first.

***

## palette-auditor

**Model**: Sonnet · **Effort**: high · **Max turns**: 20

Specialized accessibility and quality auditor. Reads pre-computed WCAG and APCA scores from palette data — it never recalculates contrast from scratch.

**Responsibilities**:

* WCAG 2.1 and APCA contrast audits across all shade/text color pairs
* Global contrast score and risk level classification
* Risky pair detection (fails AA, fails large text, borderline)
* Actionable remediation recommendations (lightness adjustments, swap suggestions)

***

## palette-codegen

**Model**: Sonnet · **Effort**: high · **Max turns**: 12

Specialized code and token export agent. Minimal questions, direct output.

**Responsibilities**:

* Call `generate_code` with the right `format` and `colorSpace`
* Optionally commit the generated file to the project repository
* Handle both primitive-only and semantic (primitives + semantics) exports

***

## palette-publisher

**Model**: Sonnet · **Effort**: high · **Max turns**: 18

Specialized palette lifecycle agent for the UI Color Palette platform.

**Responsibilities**:

* Browse and search community palettes (`list_published_palettes`)
* Retrieve a specific palette by ID and load it into session context
* Publish, update, share, unshare, and delete palettes
* Handle OAuth authentication flow for protected operations

***

## palette-transitioner

**Model**: Sonnet · **Effort**: high · **Max turns**: 24

Specialized transition agent. Takes `PaletteData` and moves it into the right design-tool or file artifact.

**Responsibilities**:

* Variables (Figma, Sketch, Penpot)
* Tokens (Penpot, Sketch, Framer)
* Styles (Figma, Framer, Sketch)
* Swatch board / document previews
* Semantic variable collections on top of primitive variables

It handles the transition logic — `uicper` handles the broader orchestration.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.ui-color-palette.com/claude/agents.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.