PixelLab Pip

Meet PixelLab Pip: a tiny pup who fetches the right PixelLab workflow. He follows human commands to create, edit, and animate pixel assets, then sniffs out a bigger prompt, scouts for a useful tool, and carries back what happened.

For PixelLab.ai.

Unofficial community project: PixelLab Pip is not affiliated with PixelLab.

Table Of Contents

• Features
• Install
• Usage
• Setup MCP / API
• Security
• Benchmark
• Showcase ↗
• Developer ↗
• Resources ↗
• More Documentation ↗
• MIT License ↗

Features

Use me when you need to create, edit, animate, integrate, or troubleshoot PixelLab assets and workflows.

Install

Agent-Assisted Install

Ask your local coding agent:

Install PixelLab Pip from https://github.com/Shilo/pixellab-pip using the proper marketplace or plugin workflow for this agent. Prefer marketplace install/update over copying files.
Then run `/pixellab-pip setup` or `@pixellab-pip setup`.

Marketplace Install

Install PixelLab Pip as a plugin or extension when your agent supports marketplaces. Use manual skill install only when your agent does not support plugin installation.

Claude Code

Install:

/plugin marketplace add Shilo/pixellab-pip
/plugin install pixellab-pip@pixellab-pip
/pixellab-pip setup

Update:

/plugin marketplace update pixellab-pip
/plugin update pixellab-pip

Codex

Install:

codex plugin marketplace add Shilo/pixellab-pip
codex plugin add pixellab-pip@pixellab-pip
@pixellab-pip setup

Update:

codex plugin marketplace upgrade pixellab-pip
codex plugin remove pixellab-pip
codex plugin add pixellab-pip@pixellab-pip

Cursor

Use Cursor's plugin marketplace or team marketplace flow when available, or install the raw skill manually.

/pixellab-pip setup

Gemini CLI

Install:

gemini extensions install https://github.com/Shilo/pixellab-pip
/pixellab-pip setup

Gemini uses gemini-extension.json for installation and GEMINI.md for invocation context. Both point back to the same skill contract in skills/pixellab-pip/SKILL.md.

Update:

gemini extensions update pixellab-pip

GitHub Copilot CLI

Install:

copilot plugin marketplace add Shilo/pixellab-pip
copilot plugin install pixellab-pip@pixellab-pip
/pixellab-pip setup

Update:

copilot plugin update pixellab-pip

VS Code Agent Plugins

Use Chat: Install Plugin From Source with this repo URL or VS Code's plugin marketplace flow.

/pixellab-pip setup

OpenCode

OpenCode discovers Agent Skills natively — there is no marketplace step (its opencode plugin command installs JavaScript code plugins, not skills). Install the skill (see Manual Skill Install) into any path OpenCode reads: .opencode/skills/, .agents/skills/, or .claude/skills/ (project, or their ~/.config/opencode, ~/.agents, ~/.claude globals). OpenCode loads it on demand through its built-in skill tool.

.opencode/skills/pixellab-pip/SKILL.md

Then ask OpenCode to set up PixelLab Pip.

Deep Code (DeepSeek V4)

Deep Code — the DeepSeek V4 terminal agent — discovers Agent Skills natively; there is no marketplace. Install the CLI, then drop the skill into a path it reads — ~/.agents/skills/ for every project, or .agents/skills/ or .deepcode/skills/ inside one project:

npm install -g @vegamo/deepcode-cli

.agents/skills/pixellab-pip/SKILL.md

Open Deep Code's skill picker with / or type the skill name, then run PixelLab Pip setup.

Manual Skill Install

Manual install is useful for project-local setup or agent apps that support raw Agent Skills without plugin installation. Copy the contents of skills/pixellab-pip/ into a folder named pixellab-pip inside your agent's skills directory, so SKILL.md is directly inside the final folder.

You can also download the skill zip from the latest release and extract it into your agent's skills directory.

.agents/skills/pixellab-pip/SKILL.md
.claude/skills/pixellab-pip/SKILL.md
.cursor/skills/pixellab-pip/SKILL.md
.opencode/skills/pixellab-pip/SKILL.md

PowerShell:

New-Item -ItemType Directory -Force .agents\skills\pixellab-pip
Copy-Item -Recurse -Force skills\pixellab-pip\* .agents\skills\pixellab-pip\
# Then run in Codex: $pixellab-pip setup

macOS/Linux shell:

mkdir -p .agents/skills/pixellab-pip
cp -R skills/pixellab-pip/. .agents/skills/pixellab-pip/
# Then run in Codex: $pixellab-pip setup

Use .claude/skills/pixellab-pip or .cursor/skills/pixellab-pip instead if that is the skill directory your assistant reads.

Usage

Use Pip with a normal prompt:

/pixellab-pip <prompt>
@pixellab-pip <prompt>
$pixellab-pip <prompt>
pip <prompt>

The plain pip form can work as an implicit skill call because Pip's metadata describes PixelLab and Pip intents. Explicit /pixellab-pip invocation is still recommended when you want this skill used for sure.

Examples:

/pixellab-pip make a cute knight character sprite
pip animate this idle character
@pixellab-pip edit this image into cleaner 32px pixel art
@pixellab-pip create a tiny potion sprite, then make an Aseprite workspace with frames, a tag, a GIF preview, and a sprite sheet
@pixellab-pip create a mossy top-down tileset, then clamp the result to 1-bit black and white
@pixellab-pip make a small item icon sheet and reduce it to a Game Boy green palette

Implicit invocation should also work when an agent sees PixelLab/Pip context plus setup or asset words such as "setup PixelLab", "configure PixelLab MCP", "connect the PixelLab API", "create an image", "make a sprite", "draw a character", "generate a tileset", "animate this", "edit this image", "use PixelLab MCP", "call the REST API", or "check PixelLab docs". Explicit invocation is still recommended when you want Pip used for sure.

Commands

Setup

/pixellab-pip setup
@pixellab-pip setup
$pixellab-pip setup

Runs the beginner-friendly PixelLab setup wizard. Pip recommends MCP + API for AI assistants/editors, can configure documented REST v2 fallback for Pip with the same PixelLab Secret, and only changes settings after a token-free preview and explicit approval.

Bark

/pixellab-pip bark
@pixellab-pip bark
$pixellab-pip bark
/pixellab-pip bark on
/pixellab-pip bark off

Toggles Pip's completion sound. Bark is on by default and persists in skills/pixellab-pip/pixellab-pip.json next to SKILL.md when the installed skill directory is writable, with an exact user-config fallback only if skill-local persistence fails. Because bark starts on, a first-run bark command usually toggles it off without playing; use bark on to test the sound. Pip only barks after a live PixelLab generation, edit, transform, conversion, background-removal, or animation job finishes successfully. It does not bark for setup, auth checks, balance/status checks, docs, failed or pending jobs, downloads alone, or local post-processing alone.

The current sound is hardcoded in the bundled helper as skills/pixellab-pip/assets/bark.wav. Running bark or bark on plays the sound immediately when bark ends up on, which doubles as a sound test. If the helper cannot run, Pip falls back to a native system success or alert sound instead of passing an audio file path to the app.

Setup MCP / API

For most users, run /pixellab-pip setup and choose MCP + API. MCP connects PixelLab directly to your AI assistant/editor/app, and API fallback means documented REST v2 fallback when MCP tools are unavailable, incomplete, or insufficient.

Bundled helper scripts require Python 3.10 or newer when your agent runs local tooling such as background-removal validation, bark configuration, docs-drift checks, or release QA.

PixelLab generation requires a PixelLab bearer token and may spend credits. PixelLab Pip does not include, store, or print that token. Do not use copied website session tokens or undocumented internal endpoints used by first-party surfaces such as the website or Aseprite extension for automation unless they appear in PixelLab's public REST v2 docs/OpenAPI or MCP docs as supported programmatic endpoints/tools.

Open the PixelLab account page after signing in and copy the value labeled Secret. PixelLab may call this value an API key, API token, secret, or token. Pip calls it a bearer token for MCP and REST v2 auth. Store the token outside chat. The recommended local secret name for Pip REST v2 fallback is:

PIXELLAB_SECRET

Never paste the Secret into chat, commit it, print it in logs, copy browser session tokens, or ask an agent to scan .env*, shell history, home directories, or environment dumps. For deeper setup, auth, and service-boundary details, see More Documentation ↗ and PixelLab Auth And Security ↗.

Security

PixelLab Pip is safe to install — and checked by outside services, not just by us. It is plain-Markdown instructions plus two tiny local Python helpers (a completion sound and a background-removal check): no compiled binaries, no hidden network calls, and it never reads or prints your PixelLab token. Every release is independently scanned and cryptographically signed. See any of it for yourself:

• Skill audit — NVIDIA SkillSpector (the same engine ClawHub's registry uses) scans the skill on every push and weekly. Results are public on the Code Scanning tab, and a release is blocked if it goes over the risk threshold.

• Independent registry audit — every release auto-publishes to ClawHub, whose own third-party audit (SkillSpector + VirusTotal + ClawScan) posts a public verdict: view the ClawHub security audit ↗.

• Malware scan — each release links a public VirusTotal report of the exact download in its release notes.

• Tamper-proof build — every release zip comes with a cryptographic proof-of-origin (a Sigstore build-provenance attestation): confirmation that GitHub built this exact file from the public repo and no one altered it afterward. Anyone can check a downloaded zip:

  gh attestation verify pixellab-pip-<version>.zip --repo Shilo/pixellab-pip
  

• Report a concern — see SECURITY.md; private vulnerability reporting is enabled on this repo.

What the skill uses: your PixelLab token — only as an auth header for PixelLab requests. Your MCP client passes it (or REST fallback references PIXELLAB_SECRET by name); Pip is designed never to read the token's value into the conversation, print it, log it, or store it. The skill also uses the python command for two small bundled helpers (completion sound and a background-removal check, which make no network calls of their own) and writes only to your project's pixellab-pip-generations/ output folder and the skill's own config file. It never scans your other secrets, .env files, shell history, or browser sessions — and it works for guidance, setup, and routing with no token at all.

Expected scan disclosures: Pip legitimately documents PixelLab bearer-token handling, official api.pixellab.ai documentation URLs, and a local sound-playback helper, so instruction scanners flag those as by-design findings. Each is dismissed on the Code Scanning tab with a public, per-finding rationale you can read. No scanner can prove an instruction file safe; these are layered, independently verifiable checks, not a guarantee.

Benchmark

Reproducible measurement of what the skill costs an agent and how well it routes, versus the alternatives. Produced by dev-tools/skill_benchmark.py — it measures the agent session (context tokens, output, cost, routing correctness), never PixelLab credits.

Across 20 scenarios phrased as plain human requests (claude + codex + deepseek-v4-pro, 1 rep) — no tool names or "which endpoint" hints, so the only difference between arms is what that user would actually have: the skill, the docs link's content, or nothing — each agent plans the route and is scored only when it identifies the correct PixelLab route (no partial credit for a plausible-sounding plan that names the wrong tool; the table averages per-scenario check rates across the three agents). The skill routed every scenario correctly across all three agents (the only dropped sub-check is the exact credit figure on one cost question); the official docs got ~58% — strong on pure MCP tool-selection but 0 on every REST-only route (backgrounds, image-pro sheets, the skeleton flow) and near-0 on setup, with mixed credit on blueprint replay; no skill was mostly lost. The skill starts slightly leaner than the docs up front (~7.5k vs ~7.7k) but reads a reference when a task calls for one (11 of these 20 scenarios), averaging ~9.1k per session — the routing accuracy costs more context than the docs on reference-needing tasks. The benchmark is dry — it spends no PixelLab credits. A dated, nondeterministic snapshot; see the full per-scenario tables and reproduce steps in the full benchmark report ↗.