svg-picker

A Human-in-the-Loop Icon Selector for AI Agents

Inspired by HULA: Human-In-the-Loop Software Development Agents (ICSE SEIP 2025)

What Is This?

svg-picker is a lightweight tool that gives AI coding agents the ability to ask a human to visually select an icon, then returns the SVG source code directly to the agent via stdout.

Instead of guessing icon names or embedding random SVG strings, the AI can now say:

"I need a home icon — please select one from the options"

A native dark-themed window opens. The human picks. The AI gets clean SVG code.

The human provides visual judgment. The AI handles everything else.

Background: Why "Human-in-the-Loop"?

The paper HULA: Human-In-the-Loop Software Development Agents (Takerngsaksiri et al., ICSE SEIP 2025) demonstrates a critical insight:

Existing LLM-based coding agents rarely incorporate human feedback at intermediate stages. When humans can intervene during plan generation and code writing — not just review final output — development time and effort decrease significantly.

HULA was deployed at Atlassian JIRA and evaluated by real engineers. The results confirm what many suspected: fully autonomous AI agents produce better outcomes when humans stay in the loop.

svg-picker applies this principle to a specific, narrow task: icon selection. It is a minimal, production-ready implementation of human-in-the-loop tooling for AI agents — without the complexity of a full framework.

How It Works

User/CI:  svg-picker <keyword>
            │
            ▼
    ┌─────────────────┐
    │  Search Iconify  │   (Background thread)
    │  Download SVGs    │
    └────────┬────────┘
               │
               ▼
    ┌─────────────────────┐
    │   GUI Window        │   ← Human picks icon(s)
    │   Dark theme        │
    │   Click to select   │
    └────────┬───────────┘
               │
               ▼
         SVG code to stdout  →  AI agent consumes it

The human acts as a visual judge — the AI remains in full control of the workflow.

Features

• Native GUI — PySide6 window, no browser required
• Themable — cream / sky / dark backgrounds via --theme
• Pagination — 10 icons per page, flip with ‹ / › buttons or ← / → keys; selections persist across pages
• Iconify API — Access to 150+ icon sets, 500,000+ icons
• stdout output — SVG code flows directly into the AI's context
• Cancellation signal — closing the window writes [svg-picker] cancelled: ... to stderr so callers can distinguish user cancel from program crash
• One-step install — pip install, single command
• Zero config — No API keys, no servers, no infrastructure

Install

pip install svg-picker

Or for development:

pip install -e .

Usage

svg-picker <keyword> [--theme cream|sky|dark]

Options

Examples

svg-picker home                  # default cream theme
svg-picker home --theme sky      # sky blue background
svg-picker arrow -t dark         # dark theme, short flag

Default theme via .env

To set a project-wide default without typing --theme every time, edit .env in the directory you run svg-picker from. On first launch, svg-picker creates an empty .env for you if one doesn't exist — just add a line like:

# .env
SVG_PICKER_THEME=sky

Resolution order: CLI --theme → .env SVG_PICKER_THEME → cream. An invalid value in .env prints a warning to stderr and falls back to the built-in default.

Steps

1. Window opens with the first page of 10 matching icons
2. Click to select one or more; click again to deselect
3. Flip pages with ‹ / › buttons or ← / → keys — selections persist across pages
4. Press Confirm — SVG source code is printed to stdout, window closes
5. Close the window (X) to cancel — a [svg-picker] cancelled: ... line is written to stderr

For AI Agents

As a Claude Code Skill

Place this file as ~/.claude/skills/svg-picker.md:

# svg-picker

Pick SVG icons via keyword search with human visual selection.

Usage: svg-picker <keyword> [--theme cream|sky|dark]

The human selects icons from the window. SVG source code is output to stdout.
If the window is closed without confirming, a "[svg-picker] cancelled: ..."
line is written to stderr — read stderr to distinguish cancel from crash.

Programmatic Usage

import subprocess

result = subprocess.run(
    ["svg-picker", "home"],
    capture_output=True, text=True,
)

if result.returncode != 0:
    # 程序异常退出 —— stderr 会有 Python traceback
    raise RuntimeError(f"svg-picker crashed: {result.stderr}")

if "[svg-picker] cancelled" in result.stderr:
    # 用户主动关闭窗口,没点 Confirm
    if "selected but not confirmed" in result.stderr:
        # 选了但没用上 —— 尊重取消意图,不要 fallback
        print("User cancelled with selections discarded")
    else:
        print("User cancelled without selection")
else:
    # 正常完成 —— result.stdout 是 SVG 源码
    svg_code = result.stdout
    # 每段格式:<!-- iconify_id -->\n<svg>...</svg>

Comparison

Both share the same core principle: human judgment improves AI output.

Related Work

• HULA: Human-In-the-Loop Software Development Agents — ICSE SEIP 2025
• acte — Framework for GUI-like Agent Tools
• OpenUI — Open Standard for Generative UI

License

MIT