agentleFS
alanvaa06alanvaa06/award-craft/skills/craft/SKILL.md

craft

Skill1 starsChanged 3 months agoMITView on GitHub

Build a new Awwwards-level landing page end-to-end — intake (script + brand source + interview), approved design plan gate, recipe build, Higgsfield assets, mandatory verify. Use when the user wants a new landing/marketing page built ("/award-craft:craft <brief> [@docs] [--brand <path>]"). Not for elevating an existing page (use elevate).

10 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 4 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 9 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Run phases IN ORDER. One human gate.

    line 8
  2. Distinctive:

    Check and report before anything:

    line 11
  3. Distinctive:

    ALWAYS ask comprehension questions (even with docs): offer, audience, primary CTA, tone (3 adjectives), visual references/anti-references, …

    line 41
  4. Very specific:

    If docs + answers leave gaps → deep interview until PRODUCT.md writes with no holes.

    line 54
  5. Distinctive:

    If script and brand source contradict: surface it, propose a resolution, let the gate decide.

    line 56
  6. Very specific:

    If no landing script (guión) was supplied, CRAFT WRITES ONE as an intake deliverable: docs/guion.md — section list in scroll order, the …

    line 68

“Common” is measured: kept by at least 6 in 10 of the 9 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Craft — orchestrator

Run phases IN ORDER. One human gate. Never skip verify.

0. Preflight

Check and report before anything:

  • CORE (stop with install instructions if missing): gsap-skills skills available; impeccable v4+ available. Resolve its base dir once and reuse it (later phases need its scripts) — first hit of .agents/skills/impeccable, .claude/skills/impeccable, ~/.claude/skills/impeccable. Read version: from its SKILL.md frontmatter: below 4.0 → stop and say so, because v4 changed the shared DESIGN.md contract this plugin now writes against. Report the resolved path and version.
  • CORE for asset phase (continue but mark assets phase blocked): Higgsfield MCP tools reachable — also read the credit balance so the design plan's ceiling can be proposed against a real number.
  • OPTIONAL (warn only): remotion/hyperframes skills (custom video — out of scope v0.1), ffmpeg on PATH (scroll-video frames) — becomes CORE if the approved design plan's signature moment uses scroll-video (re-check at the gate; block the build of that section and report if missing).

1. Intake

Inputs (any combination): text brief; @docs (PRD, brand doc, landing script "guión" — script sections become the page structure); --brand <path> folder. Brand source resolution, in order: a) PRODUCT.md / DESIGN.md in the target repo → use them. They are SHARED with impeccable (its init writes PRODUCT.md, its document/documenter writes DESIGN.md). Read whatever is there and never clobber it: append the award-craft sections that are missing, leave the rest alone. An impeccable-written DESIGN.md is the incumbent visual world — direction treats it as a brand source with drift rules, not as an empty slot. b) External folder (--brand or ask) → IMPORT as local working copies with provenance (source path + imported date). NEVER write to the business folder — it is a source, not a destination. c) Neither exists → create via interview. ALWAYS ask comprehension questions (even with docs): offer, audience, primary CTA, tone (3 adjectives), visual references/anti-references, which script section is the climax (signature-moment candidate). One question at a time.

Comp authorization — ask it here, in the intake round, so it never becomes a second stop. Skip the question entirely when preflight marked Higgsfield unreachable (nothing to authorize) and note the skip for the report: "before the gate I can render three pictures of the first viewport so you approve a composition you have actually seen, not a text sketch — roughly 4-6 credits (comps need a text-capable model, which costs more than a photo slot; price it with get_cost and quote the real round total and the balance, never the ~1 credit photo figure). Yes or no?" Yes is the default worth recommending. Record the answer; direction runs or skips Pass 1.5 on it. This allowance is separate from the asset credit ceiling approved at the gate. If docs + answers leave gaps → deep interview until PRODUCT.md writes with no holes. Write PRODUCT.md from ${CLAUDE_PLUGIN_ROOT}/templates/PRODUCT.md.template. If script and brand source contradict: surface it, propose a resolution, let the gate decide.

Awareness stage (Schwartz) — ask which stage the traffic arrives at: Most Aware / Product Aware / Solution Aware / Pain Aware / Unaware. It sets both headline strategy and page length (see copywriting.md).

Voice of customer — during the interview, RECORD THE CLIENT'S EXACT PHRASING and any customer quotes they can supply (testimonials, reviews, support emails, sales-call language). Do not paraphrase: swiped customer wording is the strongest documented headline tactic. Store verbatim quotes in PRODUCT.md.

If no landing script (guión) was supplied, CRAFT WRITES ONE as an intake deliverable: docs/guion.md — section list in scroll order, the narrative arc (PAS: problem → ordered agitation → solution/CTA, see ${CLAUDE_PLUGIN_ROOT}/skills/direction/references/copywriting.md), a copy draft per section (headline + subhead + body intent), and the climax section marked. RULE: the narrative climax and the signature moment are THE SAME SECTION — the page's peak of tension is where the animation budget is spent.

2. Design plan + comps → GATE

Invoke skill direction. It produces the plan and, when the comp spend was authorized at intake and Higgsfield is reachable, three rendered compositions of the first viewport. Present the plan, the guión (script + copy drafts) and the three comps together. ═══ STOP for human approval — plan, script and composition decided in one stop ═══ Iterate on feedback. On approval: DESIGN.md written, the approved comp and its fidelity inventory recorded in the design plan; proceed.

3. Build

Invoke skill build-recipes with DESIGN.md + design plan + script + the approved comp when one exists (the comp changes the build order — reproduction first, motion second).

4. Assets

Invoke skill assets for every slot in the design plan. If blocked (preflight), leave placeholders and mark in report. The asset phase runs its own cost preflight against the ceiling approved with the plan and stops if it would exceed it.

5. Verify

Invoke skill verify. Mandatory — an unverified build is not done.

6. Final report

What was built (per section); the comp round (which comp was approved and why, or that it was skipped and for which reason — never let a wireframe-only approval read as an approved composition); verify score + pending; assets per slot, credits spent vs the approved ceiling, the approved video tier and any per-slot overrides used; drift applied; copy decisions (awareness stage, arc used, VoC quotes swiped); retro-sync suggestions (drift worth promoting to the brand source — the human decides and executes; this plugin never writes there).

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.