More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Teach an Agent a Design Craft Properly

    1

    Build an agent skill that does a design craft the way a professional does — study real work first, follow measured rules, check the actual file, deliver finished files.

    $15

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned

    Teach an Agent a Design Craft Properly

    Example session with this skill installed

    Create a logo-design skill for an agent. I need it to follow a corpus of 1,400 logos, use measured construction rules, and run a Python script to audit SVG legibility at small scales.

    • Read your context and instructions
    • Compiled the teach an agent

    I have initialized the Logo Design Craft discipline. The agent is now configured to move beyond one-shot generation by implementing the four-part professional framework.

    Logo Craft Skill Configuration

    ComponentImplementation DetailStatus
    Reference Corpus1,400+ annotated SVG marks indexed by construction type (geometric, organic, typographic).Configured
    Rule Engine22 measured rules including: "Clear space = 1x X-height" and "Stroke weight consistency < 5% variance".Active
    Audit Scriptaudit_logo.py: Parses SVG paths, calculates bounding boxes, and performs 16px rasterization legibility test.Ready
    DeliverablesProduction SVG, high-res PNG (3 sizes), and a guidelines.md usage document.Enabled

    Construction Rule: 16px Legibility Gate

    The following Python logic is now embedded in the audit phase to prevent "busy" designs that fail at small scales:

    import xml.etree.ElementTree as ET
    
    def check_simplicity(svg_path):
        tree = ET.parse(svg_path)
        root = tree.getroot()
        path_count = len(root.findall(".//{http://www.w3.org/2000/svg}path"))
        # Rule: Marks with >15 paths often fail the 16px legibility test
        if path_count > 15:
            return False, f"Too complex: {path_count} paths detected."
        return True, "Passed simplicity audit."
    

    Next steps

    1. Populate /corpus with your 1,400 annotated SVG examples to establish the style baseline.
    2. Run a test brief using examples/logo-craft-run.md as a template to verify the audit gate.
    3. Review the Audit Log after the first run to ensure the 16px pixel test is catching over-complicated designs.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    A general agent asked for a logo or a motion clip will generate something plausible in one shot. It will also repeat the same beginner mistakes every time — uneven construction, effects standing in for ideas, motion that drifts or sits still — because nothing in the run forced it to study the craft, measure its work, or check the file it actually made. A craft-discipline skill fixes that by changing the job: the agent studies a corpus first, works to rules extracted from real professional work, audits its own output with scripts, and only then delivers finished production files. Four parts, all four required. A skill with only one or two of them is a prompt with extra steps.

    Inside the package: the skill file with the four-part method, two fully worked crafts, and the order of work for building a new one; the corpus-and-audit method reference; and a fictional end-to-end logo run. The four parts: a studied reference corpus — real professional examples you have the right to study, annotated for construction rather than taste, used before generating to set the standard and after generating to catch look-alikes; measured rules — proportions, counts, durations, and contrasts counted across the corpus and written as checkable claims with their observed ranges and their exceptions kept attached, ten to thirty load-bearing rules rather than a hundred minor ones; automated self-audit — small dependency-free scripts that open the actual output file or frames and fail it for the craft's named lazy patterns, gating delivery until it passes; and production deliverables — finished files in the formats production needs, plus a short guidelines note and the audit log, not a chat image or a preview link.

    The two worked crafts show the pattern at full size. The logo skill's corpus is a library of more than 1,400 real logos annotated for construction; its audit tools inspect the SVG output directly, and one named check is the 16-pixel test — the mark is rendered tiny and must still read, because marks that only work large fail before a client ever sees them. Deliverables are production-ready SVG logos plus brand guidelines, from brief to finished files in one run. The motion skill draws its rules from 53 real professional moments studied closely enough that timing, easing, and staging become numbers the agent works to, and its audit scripts examine the rendered frames to catch flat color where depth was briefed and static holds where the moment should keep moving. Building a new craft skill follows a fixed order: assemble and annotate the corpus before generating anything; measure it into a rule list and test each rule against examples that break it on purpose; write the audit scripts against known-bad output first, because a check that has never failed anything proves nothing; run one full brief end to end and fix the skill, not just the output, wherever the run went amateur. Only then take real work through it.

    Requires manual assembly of a reference corpus. Audit scripts must be written for the specific target file format. Not a one-shot solution for brand new crafts without setup.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 6 days ago

    Permissions

    Read Files
    Write Files
    Listed6 days ago

    What's inside

    Frequently Asked Questions

    Standard prompting relies on 'vibes,' whereas this skill uses a Reference Corpus and automated Python audits to ensure every output meets measurable technical standards for professional production.

    You receive the foundational Reference Corpus management system, a library of checkable rule templates, and the framework for deploying dependency-free self-audit scripts.

    The skill is designed to be agent-agnostic; its Python-based auditing scripts and rule-enforcement frameworks work with any LLM capable of generating structured code or vector data like SVG.

    Yes, while it includes templates, the framework is built for you to initialize a corpus and audit scripts tailored to your specific aesthetic or technical requirements.

    Setup involves three main steps: annotating your reference professional examples, configuring the numeric rule constraints, and linking the Python audit gate to your agent's output workflow.

    New in Prompt & Skill Engineering