agentsclimarketplace

Creative explore

Skill DKeken/codex-skills-alternative/skills/creative-explore

17 vendor-neutral Agent Skills (SKILL.md) reimplementing the Codex Creative Production + Product Design plugin workflows. Works with Claude Code, opencode, Cursor, Codex CLI, Gemini CLI, Copilot.

Install
npx -y skills add DKeken/codex-skills-alternative --skill creative-explore

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 13 stars13 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Pick a visual direction before generating any creative. Produces 3-6 named, distinct visual directions (e.g. premium minimal, bold maximalist, lifestyle/editorial, retro, techno-brutalist) with rationale + sample prompts, then generates one reference image per direction so the user can choose. Use when the user says "explore visual direction", "what style should this be", "give me directions/looks", or before ads/scene/shot work when no style is locked. Part of the Creative Production set.

SKILL.md

5.0 KB, as published. Nobody here has run it

Creative — Explore (Visual Direction)

Converge on ONE visual direction before spending generation budget on variations. Picking a direction early is the single highest-leverage creative decision — everything downstream (scenes, shots, ads, polish) inherits it. Never jump straight to final assets without a locked direction.

When to use

  • Starting a new brand, campaign, or product visual from scratch.
  • The user has a product but no agreed "look".
  • Stakeholders disagree on aesthetic and you need concrete options to react to.
  • Before creative-scene, creative-shot, creative-ads-explorer, or creative-offer when no direction is locked.

When NOT to use

  • A direction/brand guide already exists → skip to the asset skills.
  • You only need to finish one already-chosen image → creative-polish.
  • The user wants team-wide mood alignment with territories → creative-moodboard (broader than this).

Inputs to gather (ask only what's missing)

  • Product / brand + what it does, in one line.
  • Audience + the feeling it should evoke (trust? excitement? calm?).
  • Hard constraints: locked colors, logo, fonts, explicit do-nots.
  • Medium: web hero, paid ad, packaging, social, app store.

If the user can't answer "what feeling", propose 2-3 candidate feelings and let them pick — don't stall.

Workflow

  1. Diverge on directions. Propose 3-6 named directions. Each must differ in organizing principle, not just palette. For each: name, one-line vibe, palette, typography feel, photographic/illustration style, when it wins, when it fails.
  2. Write a reusable prompt stub per direction, structured scene/backdrop → subject → details → constraints. This stub is the DNA that downstream skills reuse for consistency.
  3. Generate ONE reference image per direction with your image-generation tool (any text-to-image backend: DALL·E, Imagen, FLUX, SD, Midjourney). Keep it small/fast (1024x1024). Save to ./creative/explore/<direction>.png.
  4. Present side by side with the rationale table. Make a recommendation tied to the audience + medium.
  5. Lock one (or fuse two). Hand the locked direction + prompt stub to the next skill.

Worked example

Request: "We sell a $180 ceramic pour-over kettle. Need a look for the site hero."

## Direction A — "Quiet Luxury"
vibe: restrained, gallery-like, lets the object breathe
palette: bone white / warm grey / single matte-black accent
type: high-contrast serif display + grotesk body
style: soft north-window light, shallow DOF, seamless backdrop
wins when: price is the message (premium, trust-first)
fails when: audience is young/playful — reads cold
stub: "seamless bone backdrop -> kettle 3/4 center -> soft window light, matte black accent, 85mm f2 -> minimal, no props, no text"
[ref: ./creative/explore/quiet-luxury.png]

## Direction B — "Warm Ritual"
vibe: hands, steam, morning — the feeling of using it
palette: amber / cream / oxblood   type: humanist serif
style: golden-hour 35mm, lifestyle, real texture
wins when: selling the experience, social-first
fails when: needs to look clinical/spec-driven
stub: "wood counter, morning light -> hands pouring with kettle, steam -> 35mm, warm grade -> candid, real"
[ref: ./creative/explore/warm-ritual.png]

Recommendation: A for the hero (price-led, trust), keep B's stub for social. → lock A, pass stub to creative-shot.

Quality bar (don't ship until)

  • Directions are genuinely distinct — a stranger could tell them apart blind.
  • Each has an explicit "wins when / fails when" so the choice is reasoned, not vibes.
  • Exactly one reference image per direction at this stage (no premature iteration).
  • The locked stub is concrete enough that another skill can reuse it verbatim.

Common pitfalls

  • Six shades of the same idea — if they share palette + style, they're one direction. Force structural difference.
  • Over-generating — one image per direction. Iterate AFTER selection, not before.
  • Inventing brand constraints the user never stated — mark every assumption explicitly.
  • Skipping the stub — without it, downstream skills drift and the set looks incoherent.

Handoff

Locked direction + prompt stub → creative-scene (lifestyle), creative-shot (PDP set), creative-ads-explorer (ad batch), creative-offer (promo), or creative-moodboard (expand to territories).

Tooling

Requires a text-to-image tool. If none is available, deliver the named directions + prompt stubs and ask the user to run them, then resume at selection.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.