Logo Semantic Fusion Agent Skill

Logo Semantic Fusion

Build one structurally shared idea. Do not force fusion when a simpler wordmark, single symbol, or broader identity system serves the brand better.

Route the references

• Read references/concept-methods.md before designing or substantially redesigning a mark.
• Read references/review-checklist.md before scoring, comparing, or accepting a visual result.
• Read references/source-notes.md only to explain provenance or resolve an ambiguous method.

Choose one mode

• Design or redesign: Run the full workflow. Preserve stated redesign invariants before exploring.
• Critique: Inspect the supplied artifact and report evidence, hard-gate failures, and the smallest useful revisions. Do not redesign unless asked.
• Compare: Apply the same brief, scales, and hard gates to every candidate. Allow reject all; do not invent a new candidate unless asked.
• Prompt-only: Produce a production prompt, constraints, and future acceptance tests. Do not generate an artifact or claim visual QA, similarity screening, or production readiness.

Run the applicability gate

Proceed with fusion concepting only when all are true:

• At least two distinct brand signals are worth encoding.
• A plausible shared contour, stroke, void, glyph skeleton, or module system can carry them.
• Fusion improves the brand idea without sacrificing required-script legibility, category trust, or recognition at the intended scale.

If any condition fails, state why and recommend a pure wordmark, single symbol, abstract mark, or identity-system route. In critique or compare mode, treat forced fusion as a finding instead of ending the review.

Establish the brief

Extract or reasonably infer:

• name, pronunciation, audience, category, promise, and personality;
• target contexts and smallest committed size;
• required scripts, must-keep elements, and forbidden motifs;
• color, background, accessibility, and delivery constraints.

State consequential assumptions. Do not improvise unfamiliar letterforms or characters: consult authoritative form references and require a fluent reader or native reviewer before calling the result production-ready.

Inventory semantics and screen prior art

Map up to three primary signals to visual structures and language structures. Describe geometry, not only nouns.

During early ideation, list provisional category clichés and mark the avoid map not screened; do not block concept generation on open-ended research. Before advancing a direction or claiming distinctiveness, review a bounded current sample of relevant-category marks and the closest structural analogues. Record sources, dates, recurring silhouettes, distinctive constructions, and unresolved gaps without implying an exhaustive search. If current screening is unavailable, keep the work not screened and make no originality claim. Treat this as design-direction screening, never trademark clearance.

Generate distinct concept families

Create at least six thumbnails across at least three families and three fusion methods. Make every family change both:

• the semantic pairing or interpretation; and
• the dominant silhouette or visual skeleton.

Count a change of method alone as a variant, not a new family. Name the shared structural move for every thumbnail. Reject pasted icons, complete-symbol overlaps, generic containers, explanation-dependent readings, and close imitation.

Prefer two signals in the core mark. Add a third only when the same geometry carries it.

Select before polishing

Score candidates from 1 to 5 on brand fit, fusion integrity, screened distinctiveness, smallest-size legibility, one-color reproducibility, and cultural/category safety. Use higher scores for better or safer outcomes.

Advance only when no score is below 3, fusion integrity and cultural/category safety are at least 4, and the mean is at least 3.8. A hard-gate failure in the review checklist overrides the score. Lock silhouette, fusion relationship, dominant angle, stroke or fill logic, and essential negative space before rendering.

Render in controlled passes

• Draw in black and white.
• Normalize weight, corners, curves, spacing, and optical balance.
• Remove or responsively replace details that fail at the smallest committed size.
• Add color only after monochrome works.
• Solve the wordmark lockup separately unless it is the fused mark.

Use image generation for isolated flat concept sheets, not precision masters. Rebuild selected geometry as clean vector paths when production precision matters.

Verify the actual artifact

Run the review checklist on the rendered output. Test required sizes, one color, reverse, grayscale, and silhouette. If the full mark fails at small sizes, create a responsive micro mark and define the switch point.

Do not report visual QA from a prompt, source file, or intention. If no artifact was inspected, state visual QA not run. Preserve locked invariants during local fixes, then rerun every applicable check.

Hand off proportionately

For production delivery, define clear space, minimum sizes, responsive variants, lockups, color variants, and basic vector/export requirements. Mark non-applicable items explicitly. For concept-only or prompt-only work, list these checks as pending rather than implying completion.

Deliver by mode

• Design or redesign: Return the brief, applicability result, semantic inventory, avoid map, family matrix, scores, selected principle, artifact, QA result, and handoff status.
• Critique: Return evidence-linked findings, hard-gate status, pass, revise, or reject, and minimal revisions.
• Compare: Return a normalized score table, hard-gate status, and a winner or reject all.
• Prompt-only: Return the prompt, negative constraints, and future acceptance tests. State that no artifact was inspected and visual QA and production readiness remain unverified.

Always distinguish design-direction similarity screening from jurisdiction- and class-specific trademark clearance.