Analyze Old Designs Skill

Create a practical, evidence-based social design guideline from the curated analysis of the brand's historical design boards. The approved Markdown becomes design_guidelines.md and is used by design_brief.

Inputs and evidence

• input.brand, input.locale, input.detail_level.
• context contains curated visual observations collected from sources/old_designes/, including the source PDF names and known gaps.
• Treat the context as evidence derived from supplied designs, not as an official brand book.
• Do not claim that sampled colours, inferred typography or fallback fonts are official unless the context explicitly confirms that.
• Never invent an exact font, logo rule, product asset or Figma style.

Required sections

1. Status and scope — derived vs. official, intended channels.
2. Board preparation — FB FEED, IG FEED, STORY-isti; date, content, marker, topic card and notes.
3. Visual system — palette with confidence labels, typography, spacing, composition and photography.
4. Frame patterns — stable IDs V1, V2, … with purpose and construction.
5. Sequence recipes — static feed, carousel, stories and reels.
6. Text hierarchy and density.
7. Compliance treatment — defer exact wording to active brand_info.
8. Format rules — defer current technical limits to active meta_formats.
9. Do / do not.
10. Evidence reviewed and known gaps.

Structured data

Return matching structured data so future tooling can select patterns without parsing prose:

• tokens: colours with value, usage, confidence (observed, approximate, official);
• typography: role, weight and known/unknown family;
• patterns: stable ID, name, applicable formats, construction;
• sequences: format and ordered pattern IDs;
• mandatory_elements;
• known_gaps.

Revisions

When feedback is present, revise the previous document. Keep evidence-backed parts that were not challenged. If feedback asks for an unsupported official token or asset, retain it as a known gap and add a warning.

Language

JSON keys stay English. Human-readable output uses input.locale (Croatian for hr-HR, Slovenian for sl-SI). Names of Figma frames and internal board labels stay verbatim.

Output (JSON only)

{
  "markdown": "# Design guidelines — <brand>\n\n> Status: ...\n\n## Board preparation\n...\n\n## Visual system\n...\n\n## Frame patterns\n- V1 — ...\n\n## Sequence recipes\n...\n\n## Compliance\n...\n\n## Do / do not\n...\n\n## Known gaps\n...",
  "data": {
    "tokens": [
      {
        "name": "consumer_teal_dark",
        "value": "#0A8391",
        "usage": "gradient edge",
        "confidence": "approximate"
      }
    ],
    "typography": [
      {
        "role": "headline",
        "family": null,
        "fallback": "Inter",
        "weight": "bold",
        "confidence": "observed"
      }
    ],
    "patterns": [
      {
        "id": "V1",
        "name": "Lifestyle/product hero",
        "formats": ["feed", "stories", "reels"],
        "construction": ["approved photo", "headline", "subline", "HW footer"]
      }
    ],
    "sequences": [
      {
        "format": "stories",
        "pattern_ids": ["V1", "V2", "V4", "V5", "V6"]
      }
    ],
    "mandatory_elements": ["approved risk copy", "adult-only treatment"],
    "known_gaps": ["Exact proprietary font files"]
  },
  "sources": [],
  "warnings": []
}

input.detail_level: short = essentials, standard = complete working guide, detailed = full evidence and every pattern. Never remove known gaps.