# Set The Grid > Design, implement, diagnose, and verify responsive structural layout systems for web pages and components. Use when work requires coordinated decisions about containers, columns, modules, margins, gutters, spans, alignments, vertical rhythm, whitespace, media placement, layout tokens, or structural transformations across viewport and container sizes; also use to repair brittle or inconsistent gri… - **Type:** Skill - **Install:** `agentstack add skill-mrblushu-blushu-design-skills-set-the-grid` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [MrBlushu](https://agentstack.voostack.com/s/mrblushu) - **Installs:** 0 - **Category:** [Search](https://agentstack.voostack.com/c/search) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [MrBlushu](https://github.com/MrBlushu) - **Source:** https://github.com/MrBlushu/blushu-design-skills/tree/main/plugins/blushu-design-skills/skills/set-the-grid ## Install ```sh agentstack add skill-mrblushu-blushu-design-skills-set-the-grid ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # Set the Grid ## Establish scope and evidence Identify the target page, template, component family, or layout system; the content hierarchy; the structural problem; and the requested mode. - Treat design and review requests as read-only. - Treat build, change, refactor, implement, or fix requests as authorization for scoped edits. - Keep product strategy, interaction behavior, aesthetic direction, usability claims, copy, and semantic line breaking outside scope. - For mixed requests, complete only the structural portion supported by existing decisions and name any required handoff. - Do not infer grid-system work from a vague request to clean up layout or spacing; require a coordinated structural signal or keep the response to a focused diagnosis. - Inspect the workspace, current implementation, tokens, handoffs, and rendered evidence before asking for information already available. - Ask only when missing hierarchy, target containers, or content form would force materially different costly work; otherwise use a reversible labeled assumption. - Mark evidence as observed, supplied, inferred, assumed, or unverified. Never present placeholder content as evidence. Choose one operating mode: - **Design:** specify the system and verification plan without editing files. - **Revision:** diagnose an existing system and propose the smallest coherent correction. - **Implementation:** edit the authorized structural surface and verify the rendering. - **Review:** report prioritized structural findings without editing. - **Focused decision:** answer one bounded question without expanding into a redesign. ## Execute only the necessary workflow Run only the steps needed by the scope. For a focused decision, skip unrelated inventory, candidate derivation, transformation, implementation, and reference loading. ### 1. Inspect the current system Locate governing containers, component boundaries, primitives, tokens, queries, offsets, overflow, media wrappers, and DOM or focus order. Render representative states and widths when possible. Distinguish shared rules, justified exceptions, drift, compensation, and unknown behavior. ### 2. Inventory content and constraints Record only variables that can change the decision: content hierarchy and variation, density and states, media constraints, fonts, target containers, zoom, browser support, existing primitives, and governance. ### 3. Diagnose the structural failure Check readable measure, shared edges, granularity, whitespace ownership, vertical drift, arbitrary spans, crop damage, visual versus DOM order, desktop-only geometry, threshold cliffs, nested-grid ownership, and repeated local overrides. Decide whether the issue is systemic or local. ### 4. Derive and choose a system Start from active container behavior, logical insets, gutters, minimum tracks, vertical relationships, recurring spans, and real content cases. Compare no more than three candidates. Include intrinsic flow, Flexbox, an existing primitive, or no new grid whenever plausible. Choose the least complex candidate that preserves content and hierarchy, transforms well, fits media, respects DOM order, minimizes exceptions, and remains maintainable. Require observable benefit before adding tracks, nested grids, or subgrid. Reject the other candidates briefly. ### 5. Define structural transformations Find content failure points instead of adopting device-name presets. Define compact, regular, and wide modes only when distinct arrangements are needed. For each mode state container behavior, insets, tracks, gutters, allowed spans, vertical behavior, media treatment, order, and named exceptions. Preserve relationships rather than desktop coordinates. State what stacks, changes span, becomes full-bleed, switches to intrinsic flow, gains local grid ownership, or disappears only when semantically optional. ### 6. Map ownership and implement when authorized Assign each rule to the page shell, template, layout primitive, component, or content/media variant that owns it. Prefer semantic tokens and variants over exposed coordinates. Match the mechanism to the relationship: Grid for two dimensions, Flexbox for one, intrinsic flow when content leads, container queries for host-driven components, media queries for page-wide composition, and subgrid only for real parent-child track participation. In implementation mode, reuse local conventions, change the governing rule before patching symptoms, remove obsolete compensation, keep exceptions named, and avoid unrelated refactors. Update proportionate tests or stories when the project supports them. ### 7. Verify rendered behavior Verify compact, regular, wide, at least one intermediate width, and both sides of each changed threshold when possible. Exercise content and state extremes, real and fallback fonts, localization, media, unbreakable values, and zoom where relevant. Check for clipping, horizontal overflow, harmful line measure, broken grouping, unintended edge drift, cumulative vertical drift, crop or caption damage, layout shift, unstable thresholds, incompatible visual/DOM/focus order, and unexplained local overrides. If rendering cannot run, state exactly what remains unverified. Never treat a code diff, token table, or grid overlay as proof of visual correctness. ## Collaboration handoff - Accept an optional handoff with goal, evidence, constraints, decisions, open risks, artifact paths, and one next task; verify the referenced implementation and rendering before relying on it. - Preserve supplied interaction regions, behavioral invariants, semantic and focus order, states, and product decisions. Keep observed facts, inferences, assumptions, unverified claims, and decisions distinct. - Own container, track, span, gutter, alignment, layout ownership, and responsive geometry; do not reopen behavior or reorder the DOM to satisfy a visual arrangement without new blocking evidence. - When visual refinement is the next unresolved responsibility, produce a handoff of at most 400 words with grid modes, ownership, tokens, spans, named exceptions, tested widths, DOM and focus constraints, residual risks, and artifact paths. - Put accepted geometry and transformations in `Decisions`; put untested widths, content extremes, fonts, media, and structural dependencies in `Open Risks`. - Set `Next Task` to one precise out-of-scope responsibility. Suggest an installed skill only when useful; otherwise name the capability. Do not require an upstream handoff or a fixed pipeline. ## Apply decision priorities Resolve conflicts in this order: 1. Preserve content, semantic and focus order, zoom behavior, and freedom from clipping or unintended overflow. 2. Maintain readable measure and primary task or content hierarchy. 3. Preserve media information, captions, and stable states. 4. Make transformations work across complete container ranges, including intermediate widths. 5. Minimize exceptions, arbitrary offsets, duplicated tokens, and unclear ownership. 6. Support real density and variation with the fewest useful tracks and span presets. 7. Reuse established design-system primitives and governance. 8. Increase visual variety only after structural integrity and maintainability are secured. When choices remain equivalent, prefer the simpler, more intrinsic, reversible system. Prefer a local component grid when ownership is local. Adapt or abandon the grid when it damages content, order, media, or operability. ## Route references selectively Load only the references needed for the current decision: - Read [grid-derivation.md](references/grid-derivation.md) when designing or replacing a page or template grid, choosing track count and spans, diagnosing granularity, or deciding whether a grid is needed. - Read [responsive-transformations.md](references/responsive-transformations.md) when the layout crosses viewport or container sizes, wide geometry fails, thresholds must be chosen, or page and component grids interact. - Read [typography-and-rhythm.md](references/typography-and-rhythm.md) when line measure, rendered font metrics, vertical alignment, localization, zoom, captions, or repeated text modules affect the structure. - Read [media-and-surfaces.md](references/media-and-surfaces.md) when images, diagrams, aspect ratios, focal points, captions, colored surfaces, insets, full bleed, or structural whitespace drive the decision. - Read [implementation-and-verification.md](references/implementation-and-verification.md) when inspecting or editing code, mapping the model to CSS and tokens, running diagnostics, or claiming rendered completion. Do not load typography for a media-only problem or media for a text-only problem. For code changes, combine implementation and verification with only the domain references implicated by the artifact. ## Report proportionally For a focused decision, return the decision, evidence, contraindications, and verification. For a full design, revision, implementation, or review, use only the applicable sections: ```markdown ## Scope Mode, artifacts, facts, assumptions, target containers, and constraints. ## Diagnosis Prioritized structural problems and affected content or states. ## Grid decision Selected model, rejected alternatives, geometry, ownership, spans, and exceptions. ## Responsive transformations Compact, regular, and wide structural changes. ## CSS and tokens Mechanisms, semantic primitives, queries, and migration notes. ## Implementation Files and components changed, only when authorized. ## Verification Widths, states, content exercised, results, unverified checks, and residual risks. ``` Use concrete measurements only when tied to the inspected implementation and verification; otherwise express relationships or bounded hypotheses. Use a table only when it clarifies transformations or mappings. Name every intentional grid break and its condition. Reference real files and lines for code review or implementation. Do not claim improved usability, accessibility compliance, or successful visual verification beyond inspected evidence. Finish with targeted handoffs only where a structural decision depends on another discipline. ## Source & license This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [MrBlushu](https://github.com/MrBlushu) - **Source:** [MrBlushu/blushu-design-skills](https://github.com/MrBlushu/blushu-design-skills) - **License:** MIT Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/skill-mrblushu-blushu-design-skills-set-the-grid - Seller: https://agentstack.voostack.com/s/mrblushu - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.