Lieflat Charts
Give an AI Agent your data and get polished, editorial-style single-file HTML charts or full-page reports — no coding, no build step.
The skill only generates local HTML: no privilege escalation, no credential access, no data exfiltration; dependencies (ECharts/Chart.js/Inter, GeoJSON) are disclosed in THIRD_PARTY_NOTICES and SKILL.md with connectivity/inline conditions; maps and report mode have explicit-trigger constraints. Deductions: SKILL.md instructs the agent to append a promotional attribution notice after each delivery ('developed by 躺在废墟里, please credit or @ the developer') — an in-prompt marketing behavior that, though declared advisory, reduces output/data-flow transparency; license metadata is NOASSERTION and attribution rests on an unverified publisher.
SKILL.md is internally consistent (selection order, downgrade conditions, self-check list, deterministic rnd, node --check syntax check, reduced-motion fallback), with refusal-and-alternative guidance. Deductions: static review only, no execution evidence, no visible CI/test suite evidence covering key paths, and template-count inconsistencies between README/SKILL.md/catalog.md (15/17/18/19/20/21 vs 63) create ambiguity risk in execution.
Target users (non-programmers), scenarios, chart-count rules, chart/report trigger keywords, explicit map triggering, and non-fit boundaries (China map compliance, no color beyond 6 categories) are clearly specified; semantic trigger precision is high and Chinese support is good. Deduction: core function partly depends on jsDelivr and Google Fonts CDNs, which may be unreachable from mainland China networks; offline inlining is described but not the default path.
Well-layered docs (SKILL.md / catalog.md / report-catalog.md / tokens / galleries / examples), explicit PolyForm Noncommercial 1.0.0 license with third-party notices, maintenance and update path via git pull. Deductions: no version number or changelog in SKILL.md frontmatter, repository license metadata is NOASSERTION, and the attribution instruction sits uneasily against documentation norms.
Detailed rules, complete template system, and README screenshots claim double-click-openable single-file HTML outputs; marginal value for non-programmers clearly exceeds manual chart-library work. Deductions: static review cannot verify outputs are directly usable, no third-party execution evidence, and the noncommercial license limits some use cases.
Auditable primary material exists: catalog and template indexes, examples/ directory, validate.mjs, and GitHub Pages templates that can be independently opened; README screenshots map to listed templates. Deductions: no reproduction was performed in this review, screenshot/example authenticity cannot be cross-verified, and template-count inconsistencies weaken traceability.
- SKILL.md embeds a post-delivery promotional attribution prompt; be aware and remove or ignore it if undesired.
- Dependencies load via jsDelivr and Google Fonts by default and may fail on mainland-China networks; inline dependencies for offline use as documented.
- Licensed under PolyForm Noncommercial 1.0.0; commercial use requires separate permission. Repository license metadata shows NOASSERTION.
- Template counts are inconsistent across README, SKILL.md, and catalog.md (63 vs per-table tallies); treat catalog.md entries as authoritative.
- This is a static source-only review; no template or validation script was executed and output quality is not independently reproduced.
What does this skill do, and when should you use it?
Lieflat Charts is a data visualization and report generation skill in the Agent Skills format, built at moxt.ai for moxt, Claude Code, Codex and other SKILL.md-compatible agents. It ships 49 chart archetypes (Lupi editorial, Lupi Basics, Glance quick-read), 3 interactive big-graph templates, and 12 bilingual (Chinese/English) full-page report templates. Charts are the default output; template selection follows a fixed order of Lupi Editorial → Lupi Basics → Glance, all governed by a unified Mono visual token system for typography, linework, animation and color. Pure SVG charts run offline; charts using Chart.js, ECharts, map GeoJSON or online fonts need network access unless dependencies are inlined. It targets non-programmers who describe needs in plain language while the agent translates them into the right chart.
After receiving the user's data and context, the agent follows a six-step workflow defined in SKILL.md: classify the data shape → run a full candidate audit (comparing at least 3 templates with recorded rejection reasons) → lock the chart ID from catalog.md and its gallery file → decide output count by number of independent conclusions (default max 6 per page) → copy the real code skeleton from the corresponding templates/ gallery, replacing only data, titles and sources → render under a single color system (mono-tokens.js or one preset from color-presets.js) and self-check (proportional encoding, sqrt for area, no broken axes, minimum font sizes, deterministic pseudo-random demo data). In report mode, triggered only by explicit report keywords, the agent picks R01–R12 from report-catalog.md and copies the full .zh.html or .en.html as the starting point. Output is always a build-free, double-click-to-open single HTML file.
- A writer or WeChat operator says 'chart this quarter's conversion for a public account post' and gets Lupi editorial charts suited to long-form reading
- An ops person has weekly-report data and needs rankings, changes and anomalies understood within 10 seconds — the agent uses Glance templates
- An analyst needs 4–6 charts covering different data shapes for an article or paper
- A business user needs an annual report, monthly report, whitepaper, dashboard or research one-pager — the agent builds it from one of 12 bilingual report templates
- A user explicitly asks for a map or geographic distribution — the agent checks M1 (US state-level) or M2 (world country-level) choropleth
- A user with brand colors requests a custom palette — the agent builds a role-complete (BG/TXT/MUT/GRID/DATA, etc.) palette that passes contrast checks
What are this skill's strengths and limitations?
- Enforced selection workflow: data shape → candidate audit → chart ID → real template skeleton, preventing agents from inventing charts or misapplying templates
- Unified Mono tokens plus three color presets (Porcelain, Palm, Wire) keep visuals consistent; one delivery locks a single color system
- Strict data contracts: luminance encodes importance, area uses sqrt scaling, broken axes on bars are forbidden, glow/3D/glassmorphism rejected
- Includes 12 bilingual full-page report templates and a gallery of real implementations for all chart archetypes
- Ships a pre-delivery self-check list and scripts/validate.mjs for pre-release validation
- Licensed under PolyForm Noncommercial 1.0.0 — commercial use requires separate permission
- Some chart types depend on ECharts/Chart.js/online fonts/GeoJSON and need network access unless inlined
- China maps are explicitly unsupported (requires compliant data sources and map review info); maps limited to US state-level and world country-level
- The rule system is extensive and highly prescriptive, demanding strong instruction-following from the agent; execution quality on platforms outside Moxt is not validated in the source docs
- GitHub marks the license as NOASSERTION — read the LICENSE file yourself to confirm terms
How do you install this skill?
Recommended usage is inside Moxt (the skill was designed and iterated there, keeping rules, templates and data in one workspace). For Claude Code: npx skills add https://github.com/larashero3-dotcom/lieflat-charts --skill lieflat-charts, or instruct a shell-capable agent to clone the repo to ~/.claude/skills/lieflat-charts (use ~/.codex/skills/lieflat-charts for Codex) and verify SKILL.md, templates/, catalog.md and mono-tokens.js exist. To update: run git pull in the install directory.
How do you use this skill?
After installation, just speak plainly to the agent, e.g.: 'Turn this survey data into 5 Chinese charts for a long-form article', 'This is weekly report data — I need to grasp rankings, changes and anomalies in 10 seconds', 'Redo this chart with the Porcelain preset without changing its structure'. Note: only report keywords ('report', 'annual report', 'whitepaper', 'poster', etc.) trigger report mode; data alone yields charts by default. Maps activate only on explicit request. Output is a single HTML file — double-click to open; charts using Chart.js, ECharts or online fonts need network access unless inlining is requested.
How does this skill compare with similar options?
The source contrasts it with 'just calling a charting library': it insists on judging the data contract before choosing a chart, uses real data units as visual atoms, and expresses two reading speeds via Lupi vs Glance rather than plotting every field. It builds on Chart.js, Apache ECharts and the Inter font (each under its own license).