kami

name kami description Typeset professional documents and product landing pages: resumes, one-pagers, white papers, letters, portfolios, slide decks, landing pages. Warm parchment, ink-blue accent, serif-led hierarchy. CN uses TsangerJinKai02, EN uses Charter, JA uses YuMincho (best-effort). Triggers on "做 PDF / 排版 / 一页纸 / 白皮书 / 作品集 / 简历 / PPT / slides / Marp / markdown slides / マークダウンのスライド / 落地页 / 官网 / landing page / product page", or "build me a resume / make a one-pager / design a slide deck / turn this into a PDF / make this presentable / create a landing page". kami · 紙 紙 · かみ - the paper your deliverables land on. Good content deserves good paper. One design language across documents and landing pages: warm parchment canvas, ink-blue accent, serif-led hierarchy, tight editorial rhythm. Part of Kaku · Waza · Kami - Kaku writes code, Waza drills habits, Kami delivers documents . Update check (non-blocking). At the start of a task, run bash scripts/check-update.sh . On the first run each calendar day, it writes a marker under the local XDG cache directory and then resolves Kami's latest published GitHub Release. It uploads no user document or task content. When a newer Kami is available it prints one line; relay that line to the user, then continue. It fails silently when offline, sandboxed, or without curl , and must never block the work. Step 0 · Load brand profile (if exists) Check ~/.config/kami/brand.md (preferred) or ~/.kami/brand.md (legacy fallback). If found, read references/brand-profile.md for the full four-layer application spec (placeholder substitution, session defaults, visual customization, habit notes) and its six guardrails. If no profile exists, continue without interruption. Key rule: explicit prompt > editorial judgment > habit notes > frontmatter defaults > built-in defaults. Profile fills gaps silently; it never overrides the current conversation. Step 0.5 · User project style scan (opt-in) Run this only when the user explicitly references a sibling project as a visual reference: "like my site", "match the style of ", "use the look from ". Skip silently when no such reference exists. When triggered, before generating: locate the referenced project's declared design tokens (CSS variables, tailwind/theme config, token files) and extract dominant colors, font stack, spacing scale, and border-radius scale; prefer declared tokens over inline literals. Merge into the in-session brand profile as Layer C (visual customization), not Layer B (session defaults); never override an explicit --brand flag or values the user typed this turn. Report back in one line before continuing: "scanned , extracted N colors / M fonts; using as visual reference." Skip and fall back to the brand profile defaults if the referenced path does not exist, no CSS-like files are found, or the extraction would conflict with the user's explicit values in the current message. Step 1 · Decide the language Match the user's language. Chinese -> *.html / slides-weasy.html . English -> *-en.html / slides-weasy-en.html . Japanese -> CJK path ( .html / slides-weasy.html / slides-marp.md ) as best-effort, JP Mincho first, visual QA before shipping. Korean -> dedicated *-ko.html / slides-weasy-ko.html family, with slides-marp.md as the best-effort Markdown path, visual QA before shipping. Reference docs are shared English specs. User language HTML templates Slides (PDF default) Slides (PPTX fallback) Slides (Marp) Chinese (primary) *.html slides-weasy.html slides.py slides-marp.md English *-en.html slides-weasy-en.html slides-en.py slides-marp-en.md Japanese (best-effort) *.html slides-weasy.html slides.py slides-marp.md Korean (best-effort) *-ko.html slides-weasy-ko.html n/a (use slides-en.py only if PPTX is required) slides-marp.md Other languages (best-effort) choose CJK or EN path by script coverage, then verify manually choose slides-weasy.html or slides-weasy-en.html , then verify manually use slides.py / slides-en.py only if PPTX is required choose slides-marp.md or slides-marp-en.md , then verify manually Default to the WeasyPrint HTML path; fall back to PPTX ( slides*.py ) only when the user explicitly needs an editable deck. Design, writing, production, and diagram guidance live in CHEATSHEET.md and references/*.md . Step 3's tier table decides how much of it to open; read the lowest tier that covers the task, not all of it. Code blocks with class="language-*" are highlighted only when optional Pygments is installed in the build environment. Without it, PDFs still render and code blocks stay monochrome. Mathematics is strict LaTeX, never pseudo-math. When a document contains mathematics, author formulas only as standard LaTeX delimiters in rendered body text: inline \( ... \) and display \[ ... \] . Delimiters inside title metadata, form options, literal/code, script, style, SVG, template, and MathML regions are treated as text. Do not replace formulas with Unicode approximations, ASCII fractions, screenshots, or raw TeX printed on the page. The renderer requires Node.js 20 or Node.js 22 and newer. Before shipping HTML/PDF, run bash scripts/ensure_mathjax.sh , then python3 scripts/math_render.py --in-place filled.html ; this converts every formula into MathJax SVG. render_pdf also performs this conversion in memory as a hard fallback and fails if MathJax or the TeX is invalid. Finish with python3 scripts/math_render.py --check filled.html ; a completed HTML must contain no raw TeX delimiters in rendered body text. Step 1.5 · Intent extraction (silent) Before picking a template, silently confirm purpose, audience, hard constraints, and what outcome counts as success. Skip any dimension the conversation or the document type already answers (a resume's purpose is always "get an interview"). Question budget, global for this skill: infer first, from the decision tree and context. Ask at most once, in one compact question (max 2 sub-questions), and only when 2+ intent dimensions are genuinely unresolvable or two templates genuinely both fit. Never present the dimensions as a checklist. Execution contract Before creating or modifying an output, lock the contract: language, template, output format, page or length target, visual acceptance check, and verification command. Infer from the user's request when clear; ask only when missing fields materially change the deliverable. Use the nearest existing template and verification path. Do not add a new template, shared CSS layer, dependency, script flag, or optional mode unless the current request cannot be satisfied without it. A change to SKILL.md , templates, scripts, or references reaches npx skills add and plugin installs on the next push to main ; Claude Desktop users get it only when the next release rebuilds kami.zip . Name the channel that carries the change before calling it shipped. Work mode Route by the artifact's current state before loading more guidance. This is an internal branch, not a new user-facing command. Current task Mode Contract New document or substantial restructuring New document Lock the execution contract, write content.json with brief + content , then fill and verify Text replacement, translation, or factual correction in an existing artifact Content-only Preserve CSS and layout unless the new copy proves a fit defect User supplies a render or screenshot and rejects how it looks Visual repair Treat the render as the current brief, lock target + preserve boundary, make the smallest fix, then verify the affected matrix Standalone generated illustration, cover, social card, or redraw Generated asset Lock the semantic image brief before pixels; preserve accepted parts across iterations Step 2 · Pick the document type User says Document CN template EN template KO template "one-pager / 方案 / 执行摘要 / exec summary" One-Pager one-pager.html one-pager-en.html one-pager-ko.html "white paper / 白皮书 / 长文 / 年度总结 / technical report" Long Doc long-doc.html long-doc-en.html long-doc-ko.html "formal letter / 信件 / 辞职信 / 推荐信 / memo" Letter letter.html letter-en.html letter-ko.html "portfolio / 作品集 / case studies" Portfolio portfolio.html portfolio-en.html portfolio-ko.html "resume / CV / 简历 / 履歴書" Resume resume.html resume-en.html resume-ko.html "slides / PPT / deck / 演示" Slides slides-weasy.html slides-weasy-en.html slides-weasy-ko.html "个股研报 / equity report / 估值分析 / investment memo / 股票分析" Equity Report equity-report.html equity-report-en.html equity-report-ko.html "更新日志 / changelog / release notes / 版本记录" Changelog changelog.html changelog-en.html changelog-ko.html "landing page / 落地页 / 官网 / product page / 产品页" Landing Page landing-page.html landing-page-en.html landing-page-ko.html Changelog vs. release notes : The changelog template above is for styled document output. GitHub release notes are a separate deliverable; use /write with Release Note Template Mode. Landing Page : Screen-first interactive template. No PDF output. Includes gallery carousel with auto-rotate, hero entrance animation, responsive breakpoints (880px / 480px), and prefers-reduced-motion support. Deploy as static HTML to Vercel / Netlify / any host. The agent fills {{PLACEHOLDER}} values and HTML comment blocks, then saves as a ready-to-serve .html file. Landing Page companion files : For a production multilingual deploy, copy the five landing-page-*.example files alongside the main HTML and drop the .example suffix. {{SITE_ORIGIN}} is the scheme + host of your {{CANONICAL_URL}} (e.g. https://example.com ). What each file covers, and the opt-in Accept-Language redirect: references/design.md Section 11 «Companion assets». Production product site mode : docs, help, releases, changelog, roadmap, legal pages, or more than two locales means a site system, not a page. Lock the five items below from the source material before filling any template, asking only where a missing item would change the deliverable. Then follow references/design.md Section 11 «Product site system», and «Documentation site» in the same section once the page grows a docs or help shell. # Lock 1 Product category - first-viewport category: app, CLI, terminal, utility, skill, template system, or another user-provided label. 2 Real assets - available product screenshots, logo, icon, or UI captures, mapped to hero/gallery/feature/social slots. Missing assets must stay marked, not replaced with stock imagery. 3 Site shape - single page, or home plus docs/help/releases/changelog/roadmap/legal pages? 4 Locales - exact locale list, canonical paths, and whether a generator/check mode is needed. 5 Truth surfaces - install path, price, version, support route, FAQ, llms.txt , and llms-full.txt that must stay synchronized. Keep project-specific release artifacts, payment providers, appcast rules, and private local paths out of Kami. Decision tree (use before asking) Ask only when two cells genuinely both fit. Signal Document Length target unknown Infer from the content and document type; ask only if the choice materially changes the deliverable, within the question budget ≤ 1 page + investor / recruiter / exec summary audience one-pager ≤ 1 page + formal correspondence (sales, hiring, resignation, memo) letter 1.5-2 pages + career narrative + project bullets resume 3-6 pages + project showcase + visual heavy portfolio 6-15 pages + sustained argument + low visual density long-doc Presentation flow + speaker support + per-slide assertion slides Financial / metrics dashboard + thesis + price or risk view equity-report Version-by-version log + release facts changelog Product showcase + pricing + screenshots + FAQ for browser landing-page Ambiguity examples that justify a one-liner: "1.5 page career story with heavy visuals" -> ask "resume or portfolio?" "2 page exec summary with metric tiles" -> ask "one-pager or equity-report?" "5 page argument with several charts" -> ask "long-doc or portfolio?" Diagrams (primitives, not a separate template type) When the user asks for a diagram inside a long-doc / portfolio / slide (not a standalone document), route to assets/diagrams/ rather than a template. Eighteen types ship there; references/diagrams.md section 1 «Selection» maps each to what it shows, and section 7 lists the AI-slop patterns to avoid. Read it before drawing. Extract the <svg> block from the chosen file and drop it into a <figure> inside the document. Three routes inside that file, by trigger: Trigger Section Full-system panorama, control plane, roadmap, or owner map in one artifact «Architecture boards». Do not inflate the single architecture figure past its node budget. A diagram that lives in the user's repo (README figure, "给项目画张架构图", updating an existing one) «Maintained diagram assets». Evidence pass first, keep the index.html + PNG + prompt.md trio consistent, never redraw from memory or hand-edit the PNG. A standalone raster illustration or a redraw of a photo or screenshot «Illustration briefs», plus the Illustrations rules below. Auto-select charts from data. When content carries numbers, pick the chart type yourself and embed it without asking. Two house calls that differ from the common default: a ~100% share renders as a donut only up to 6 items, and 7 or more becomes a horizontal bar; a single time series whose absolute count changes dominate (not rate) renders as bars, not a line. When several types fit, prefer the one that shows variance most clearly. Always embed inside a <figure> whose caption states the insight, not the data range. Illustrations (host image model, not inline SVG) Inline diagrams above are vector SVGs you assemble by hand. For a standalone raster illustration, or a redraw of a figure, photo, or screenshot in the Kami look, delegate the drawing to the host's own image generation. Never call an external image API or require a key; rendering is the host's job. If the running host exposes image generation, apply the brief below and render the image directly. If image generation is unavailable, output the same complete brief as text. Route from observed capability, not a remembered list of host names. Brief: first state the claim the image must communicate, its destination and smallest display size, the accepted reference it should sit beside, and what must not appear. Then apply the Kami visual system: warm parchment ( #f5f4ed ) background, never pure white; one accent only, ink blue ( #1B365D ); all else warm gray with a yellow-brown undertone, no other colors; thin single-line geometric strokes and simple flat icons; no gradients, drop shadows, or 3D; serif labels; generous whitespace, composed like a figure in a well-typeset report. Full brief skeleton, icon rules, and QC checklist: references/diagrams.md «Illustration briefs». Switch to this path (instead of enlarging a hand-assembled SVG) when a figure needs more detail than SVG assembly holds at the target display width, typically web-article figures at teaching depth (see references/diagrams.md density tiers). When one deliverable needs several generated images, drive them through a single handoff file: one line per image (slot, aspect ratio, shared style anchor, prompt, status), generate in batches of at most 5, update the status column after each batch, and check existing generated output before regenerating. The style anchor is shared by the whole batch; per-image style drift is the failure mode. Step 2.1 · Source and material pass Run this before distilling or filling when the document depends on facts or materials outside the user's draft. Skip it for personal drafts where the user already supplied everything. Source check fires when the document names a company, product, person, release date, version, funding round, metric, market fact, or technical spec. Work from primary sources, note the source and date for facts that drive the document, and ask the user when sources conflict rather than choosing silently. references/writing.md «5. Sources before phrasing» owns the detail, including which current-sounding claims are banned until checked ("latest", "recent", "new", version numbers, launch dates, financial figures). Material check fires when the subject is a company, product, project, venue, or personal brand. Confirm what makes it recognizable before layout: Need Required when Accept Logo Any branded document User file or official SVG/PNG Product image Physical product / venue / object Official image, user image, or marked gap UI screenshot App / SaaS / website / tool Current screenshot, official product image, or user capture Brand colors Branded one-pager / portfolio / deck Official value, extracted asset value, or keep kami ink-blue Fonts Only if brand typography matters Official font, close system fallback, or kami default A missing item gets a compact gap table and one question, never a generic stand-in, an approximated logo, or an invented value. references/writing.md «6. Materials serve recognition» carries the writing-side rules. Logo fallback: when the request names no logo but the brand profile has a logo path, fill the commented .brand-logo slot in one-pager / portfolio / slides-weasy per references/brand-profile.md Layer C. Expand ~ to an absolute path; if the file is missing or the template has no slot, leave it commented and render without a logo, never a broken image. An explicit logo in the current request always wins. Materials status block After the material check, output a structured status block before continuing. This is a one-shot transparency display, not a question: Materials status: - Logo: OK /absolute/path/to/client-logo.svg - Brand colors: OK #1B365D mapped to --brand - Product screenshot: MISSING (using a pure-text layout; no placeholder image) - UI screenshot: not required for this doc type Use OK , MISSING , or not required . If a required item is missing and no user input arrived, ask once with the gap table; otherwise continue silently. Step 2.5 · Distill raw content (if applicable) Auto-detect whether to distill. Do not ask the user; judge from the input. Skip distill only when the content already maps one-to-one onto the template's sections with quantified metrics in place, or the user said "use as-is" / "直接用这个". When in doubt, run distill. Distill acceptance: every factual claim, number, date, name, source, and material reference from the raw input must land either in a template section (per-doc structure in references/writing.md ) or in a gap table of what the template needs but the content lacks. Share the gap table once; never guess to fill a gap. Example gap-check: Template needs Found Missing 4 metric cards "8 years", "50-person team" 2 more quantifiable results 3-5 core projects 2 mentioned at least 1 more with outcome Materials logo file provided product screenshot source Then proceed to Step 2.6 (slides) or the layout note (all other doc types) with structured, distilled content. Persist the distilled content as a content IR (new documents) When building a new document (not a text tweak on an existing one), write the distilled result and the locked execution contract to a content.json next to your working HTML before filling: { "type" : "resume" , "lang" : "cn" , "brief" : {