Adapted from the DGF Agent Basics series — consumer-first guides for working with a personal AI agent. Pick only what you need.
> Pick only what you need. This skill is individual. This is one person's thoughts, offered as a pre-configured help, not a must. Take only the value you need from it, prune it to what fits you, and leave the rest. You do not have to use what the author thinks, and you do not have to take any other skill with it.
Summary: Make visual and creative work by pairing a third-party taste skill with whatever creative app you name, or an easy connected route, with preview first and no posting without your go-ahead. Optional.
Use this skill when the user wants to make something visual or creative. It pairs a design taste skill with whatever creative or visual app the user wants to use, then makes a task that is specific to that app. It covers no other kind of work.
This skill stands alone. An agent that knows nothing else can follow it from the startup sequence below, then the creative method. This skill works on its own, with no other skill needed, and no knowledge of any other skill assumed.
Startup and how to use
Start by getting your own personal agent. Get your own personal agent and set it up. Bring your own accounts and logins. Use nothing that belongs to anyone else.
Tell your agent what you care about in plain words. Name your business, your work, your family, your hobbies, or any topic you want it to follow. Talking is the setup. You do not have to set up a system first.
Ask for the plan first, before any work runs. Say the task, then ask for the exact steps it will take, what it needs from you, and anything it cannot do. Wait for you to approve or correct the plan. Review the plan like a checklist and fix a wrong step or add a missing step before work starts.
Check that your agent understood you, before it runs. Ask it to show back what it thinks you mean and what output it will give you. Correct its repeat back right there when it does not match what you pictured.
Work in small single steps. Count the steps first. Give one step, check it, then give the next step. A ten step run is a huge run for an agent, so break it down and keep each step to exactly one thing.
Show it once, then save it by name. This is how you create a new task or a new skill. Take over the browser and do the task once yourself while your agent pauses. You drive alone and your agent does not watch your clicks. There is no record button to press. After you hand control back, tell your agent the steps in chat and tell it to save them by name, such as save this as Fun Things. Later, say pull up Fun Things and it picks up where you left off.
Correct it in small plain words when it does it wrong. Do not start over. Say what to change and say to remember that correction. That preference sticks and it gets better for you specifically over time.
Let it keep learning your interests every day. Your Feed looks like a newspaper with articles and topics related to what you told it you care about, updated daily. Your Ideas look like a lightbulb with suggestions for things you might like. It is not mind reading. It learns from what you actually talk about.
Use easy first and deep only when you want it. Never require a deep setup to use a basic task. Keep the wording in plain benefits and do not lead with virtual machine or VM. Say it has its own browser and workspace built in, you do not have to set anything up, it keeps your logins safe, and it saves and remembers your stuff.
Pair the taste skill with the app the user wants
Ask which creative or visual app the user wants to use, before any creative work. Tell the user they can name any app they want. Do not pick an app for them, and do not limit them to a list.
Offer the easy connected route as well, in the same message. Say that if they do not have an app in mind, your agent can steer them to apps it already has a connection route for, because those are the easiest to load and set up. Name the route in plain words, such as Canva for designs and social posts, Runway for video and image generation and editing, Figma for design files, Instagram and Facebook for posting, Google Slides for presentations, and Lovable for building a website or app. Offer Google Flow as a browser route, not a connector, where the user signs in themselves in the browser.
Check the connection before promising that route. Read the connection details, check its status, and if it is not connected, share its connect link and do not commit to the outcome until the connection is complete and healthy. A connection route existing does not mean that person's your agent is connected to it, and no account from anyone else carries over. The user brings their own accounts and logins for every app they pick.
Pair two things, and keep them separate in your notes. The taste skill gives the look and the judgment. The app gives the format, the tools, and the limits. A good look in the wrong format is not a deliverable.
Use the taste skill from https://github.com/Leonxlnx/taste-skill as the design source. Read its SKILL.md before a creative build. Follow its core method in your own words. Deal a direction from a researched set, pick the direction that carries the subject's own idea, build with real pictures of the subject's world where honest pictures exist, credit found pictures, check the work in a real browser or by watching it end to end, fix the top problems, and look again. Do not copy its scripts or its library into a project unless the user asks for that install.
State plainly that this taste skill is a third party open skill. It is not the user's own TASTE product, and it is not rebranded as theirs. Credit it when you use it.
Never invent proof. Do not invent customers, quotes, reviews, press, logos, or business statistics. Sample details that fill a gap must be listed for the user to confirm before launch or posting.
Never ask for a password or a card number in chat. Tell the user to load app logins through the secure sign-in your agent shows them. If a login was not loaded securely, work with public information only and do not look inside that account.
Never post, publish, or share creative work without the user's explicit go ahead for that piece. Preview first, always.
Make the task specific to that app
Write one app specific task for each creative job, using the five part template below. Save the filled template by name when the user wants to repeat it, so the next job in that app starts from the template, not from scratch.
Do not reuse another app's template unchanged. A video app task, a design app task, and a website task have different formats, different limits, and different deliverables.
The five part task template
1. Setup questions
Ask which app, which account, and which project or workspace the work lives in.
Ask what brand already exists. Ask for colours, fonts, logo, and any style the user wants kept.
Ask what assets truly exist. Ask for the user's own photographs, product shots, videos, reference images, or an image tool they can use. Do not plan around pictures they do not have.
Ask what the piece is for, who it is for, and the one thing it must get someone to do.
Ask what limits apply. Ask for format, size, length, orientation, platform, and deadline. Record the answers before building.
2. Question of task
State the task as one question the user can answer yes or no to. Name the subject, the app, the format, and the deliverable in that question.
State what is out of scope in the same message. Name anything the task will not do, so the user does not assume it later.
Wait for the user's yes before building. A task that starts before its question is answered is a task that drifts.
3. Question step
Break the task into single steps, in order, with one question at each step that blocks the next step.
Ask the step question, get the answer, do that step, show the result, then move to the next step. Do not run several steps on an unanswered question.
Use step questions for choices that change the result, such as which direction to pick, which picture to use, which version to keep, and which wording to use. Do not ask step questions about details that do not change the result.
Show the work after each step in a form the user can actually check, such as a preview, a screenshot, a clip, or a draft file. A step with nothing to check is not finished.
4. Change
When the user wants a change, ask what exactly changes, and what stays the same. Record both.
Change only what was asked to change. Do not rebuild the whole piece for a small change, and do not change wording, colours, or format that the user did not ask to change.
After a change, show the changed piece again, side by side with what changed, in plain words. Ask whether the change is accepted before moving on.
Save an accepted change as a correction for that task template, so the next job in that app starts with it already in place.
5. Deliverables
List the deliverables before building, with file type, size, orientation, length, and where each file lives.
Deliver every item on that list, in the same message that says the work is done, with a link or an attachment for each file. A file saved on your agent's machine has not reached anyone yet.
Include a preview the user can check, such as a video preview, a screenshot at phone width and desktop width, or a draft post. Preview is the gate for posting, never an optional extra.
List anything that still needs the user, such as sample content to swap, a credit to add, a login to load, or a final approval to post. Do not call the work finished while a deliverable is missing.
Example: a video app task, filled in outline form
Setup questions: which video app, which account, which project, what reference images exist, what characters and style are locked, what format and length, and where will it be posted.
Question of task: make one episode in this app, in this format, with these characters, for this platform, yes or no.
Question step: pick the direction and idea, pick the reference images, approve each clip, approve the assembled preview, then approve the posting package. One approval per step.
Change: change only the clip, line, or image asked for, show the new clip, and ask whether it is accepted.
Deliverables: each clip file, the assembled preview file, the posting package with hook, captions, hashtags, cover text, and posting slot, and a list of anything still to confirm before posting.
Diverse, not concrete: let the user prune
Do not make this skill concrete, narrow, or one size fits all. Present a diverse set of options first, across price, style, risk, format, and approach, so different people can see themselves in it.
Let the user prune. Ask what to cut, what to keep, and what to change, and then narrow to what they kept. Pruning is the user's job, not your agent's guess. Do not prune for them before they have seen the options.
Do not lock a single choice early. Offer at least two or three meaningfully different options at each decision point, with the trade for each, unless the user already locked that choice themselves.
Keep options that do not fit one person, because they may fit another person using the same skill. A skill that only works for one kind of person is too concrete to reuse.
When the user prunes, save the pruned version by name when they want to repeat it, so their version starts from what they kept, not from the full diverse set again.
Present the result
Lead with what was made, in plain words, and what is still waiting on the user.
State what you could not verify. State when a picture was not checked, audio was not listened through, a second version was not included, or a credit is still missing.
Stop after a denial. Do not re-pitch a rejected version in new words.
Available connectors
[Canva](/opt/hatch/connectors/canva/CONNECTOR.md) for designs, social posts, presentations, posters, image generation, background removal, resizing, and export.
[Runway](/opt/hatch/connectors/runway_mcp/CONNECTOR.md) for video and image generation and editing, music, sound effects, speech, upscaling, and background removal.
[Figma](/opt/hatch/connectors/figma/CONNECTOR.md) for inspecting and editing design files and boards.
[Instagram](/opt/hatch/connectors/instagram/CONNECTOR.md) for publishing posts, stories, reels, and carousels.
[Facebook](/opt/hatch/connectors/facebook_cli/CONNECTOR.md) for posting and publishing on pages the user manages.
[Google Slides](/opt/hatch/connectors/google_slides/CONNECTOR.md) for presentations.
[Lovable](/opt/hatch/connectors/lovable/CONNECTOR.md) for building and deploying apps and websites.
Use your agent's connector and skill search, if it has one, to find other connectors and related skills.
If nothing here fits, look on the web for a maintained API or MCP server and set it up with your agent's custom connector setup, if it has one, or build a custom connector.