01 / RESEARCH
Brand research
Profiles, source facts, user settings, usable assets, and identity references.
FINCH SKILL / V2.0.0
Take a brand collaboration through to real visual delivery.
spark-brand-collab is a single-entry Agent Skill. Given two brands and a clear task boundary, it carries work through research, direction selection, design, copy, real image production, and item-level review.
Use $spark-brand-collab to complete a collaboration for “Brand A × Brand B.” Continue with your recommended direction; produce, review, and deliver the in-scope items with a delivery index.
01 / IN ONE LINE
Turn the relationship between two brands into a reviewable collaboration and visual output.
This is not six unrelated skills. Its public entry is spark-brand-collab; it reads the right internal method for the current stage.
INPUTBrands, goal, scope, constraints
PROCESSResearch → ideas → design → copy
PRODUCTIONReal images when the host supports them
DELIVERYItem files, review records, index
02 / FLOW
Six stages, one continuous run.
For a full task, the agent normally proceeds with its recommendation. A single item, copy edit, or image revision can enter at the relevant stage.
02 / IDEATION
Collaboration ideas
Actual candidates, tradeoffs, refined directions, and a recommendation.
03 / DESIGN
Unified design
A core experience, stable materialIds, item specifications, and dependencies.
04 / COPY
Campaign copy
A theme and visible item-level language tied to the current design version.
05 / PRODUCTION
Real production
Uses the host's native image capability; otherwise preserves an honest not-generated state.
06 / REVIEW
Review and delivery
Image inspection, bounded revisions, item status, real files, and unresolved gaps.
03 / USE CASES
Run it end to end, or resume just one part.
Your task determines the scope. A single-item request does not automatically become a full kit, and useful prior work is reused.
Full brand collaboration
Research through core experience, deliverables, visuals, and an index.
One poster or deliverable
Reuse existing context and complete one explicit materialId.
Continue from an existing plan
Read current design, copy, and references, then create and inspect pending visuals.
Local copy or image revision
Update only affected items instead of rebuilding candidates or the whole kit.
04 / EVIDENCE
What is published as verified, and what is not.
Briefly: the repository documents a small fictional text run plus one native image generation, inspection, and local revision.
Published validation
The repository records a fictional two-brand run with three candidates, one chosen direction, three designed deliverables and copy, plus one generated and revised concept image.
Not fully validated
Real-brand identity fidelity, full multi-item continuity, video, and prototyping should not be inferred from that validation.