adrkit-visualize
Use when the user asks to visualize an existing ADR, its deliberation tree, or the whole decision set and its relationships.
Skill by luochang212
Chalk Labs
Need help deploying adrkit-visualize?
Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/luochang212-adrkit-visualize
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md and code. Checked Nov 2, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/adrkit-visualize
What's inside
- Skill name
adrkit-visualize- Description
- Use when the user asks to visualize an existing ADR, its deliberation tree, or the whole decision set and its relationships.
- License
- MIT
- Instructions
- 34 lines in SKILL.md · view on GitHub
Files (2)
- SKILL.md
- decision-atlas.html · examples/
SKILL.md (first lines)
# ADR Kit Visualize Seeing the decision set as a picture surfaces relationships and contradictions that a file list hides. Visualize existing records only when asked. Run `adrkit list` first, then read the record or relationships relevant to the requested view. Do not create a decision or invent a deliberation tree to make a picture. - The default visualization is the decision atlas (explicit `--format html` keeps this same object): `adrkit atlas --out "<path>.html"`. It includes implemented guidance and archived history, with archived records kept visually distinct and their bodies published in full. This is the answer to "visualize this repository's ADRs" with no further qualifier. The page chrome is English by default; pass `--lang zh` when the reader uses Chinese. - For one record's deliberation tree, use `adrkit atlas --record <name>` and save stdout to a local HTML file. If no `## Deliberation` appendix exists, say so instead of fabricating one. - For the decision map alone, use `adrkit atlas --map --format html --out "<path>.html"`; for source exports, `adrkit atlas --map --format mermaid|dot|text`. For a tree's source, use `adrkit atlas --record <name> --format mermaid|text`; `--map` and `--record` are mutually exclusive. Use `--formal-only` only when requested or useful to the question. - Text and Mermaid may be enough for docs; do not generate HTML merely because the CLI supports it. ## Theme selection Folio (warm paper, editorial typography) and Signal (dark surfaces, cyan accents, compact layout) are independent themes with the same atlas capabilities. The agent owns selection; do not make the user choose a theme as a setup step. 1. Honor an explicit theme request or a clearly stated visual preference with `--theme folio` or `--theme signal`. If the user requests randomness, use `--theme random`. 2. When updating an existing atlas, preserve its concrete `data-atlas-theme` value by passing that named theme, unless the user requests another theme or a fresh random choice. 3. For a new atlas without a clear preference, omit `--theme`: the CLI default renders Signal. Do not guess a preference from the project category, the reader's identity, or the operating system; do not prompt for a theme. Omitting `--theme` renders Signal; `--theme random` rolls a fair 50/50 coin between Folio and Signal once at generation. The resulting file keeps that theme. Do not add system appearance detection, a page switch, or a runtime reroll. Theme selection is unavailable for standalone maps and record trees. Both themes retain the same record scope, complete archived bodies, pinned source links, bilingual copy, offline behavior, map gestures and bounds, relationship focus, timeline, local search, reader navigation and keyboard behavior. Absent modules and their chapter entries disappear, the chapters renumber, and legends show only symbols actually present. Insight panels use measured complete rows near 40rem with a 50rem ceiling, including on narrow screens; sparse content stays short. Empty reference directions disappear, and when both are empty the relation area and its lower divider disappear. Preserve these mechanisms when adapting either theme; change only the presentation the user requested. Both themes keep the source link in the reader header; Folio styles it as a quiet padded control and Signal as its original text link. On viewports up to 760px, both themes hide the map's on-screen legend to keep the canvas clear; map controls and interactions remain available. Desktop keeps the data-dependent legend. Both heroes fill at least the viewport below the chapter bar. Folio keeps its original title/introduction/actions/statistics/ decision-density group and proportions. On desktop the introduction and density regions share spare height, balancing whitespace around the statistics while the lower rule meets the viewport bottom; only shorter windows compress vertical spacing, without shrinking type or the chart; Signal keeps statistics at the bottom of its hero. Taller content can scroll. Display ADR identifiers, chapter numbers and question numbers as natural integers without leading zeros (ADR 3, chapter 2, QUESTION 1), including visible labels and accessible names. Date/time fields retain their standard formats. Signal's built-in hero is a shuffled desk of real decision files, not a static illustration. Generate it with `adrkit atlas --theme signal --out "<path>.html"` and preserve its shared template behavior when adapting the page: up to three same-material folder covers, desktop landscape/mobile near-square proportions, irregular overlapping angles and depth-specific contact shadows, top-file extraction followed by reader expansion, replenishment without duplicate files, and a quiet reset only after exhaustion that returns all files and reshuffles. Use the target repository's records and counts; zero, one or two records must remain honest. Desktop hero picks open the reader on the left with the ordinary dimming/blur backdrop over the right-hand desk. Clicking that backdrop only closes the reader; opening the next file takes a separate card click. Other entry points and mobile keep the right reader. Keep all four statistics inside the hero, no bottom arrow, and automatic entrance effects confined to the hero; respect reduced motion. The runtime file shuffle is independent of the one-time choice of atlas theme. Do not replace this mechanism with decorative sample cards. ## Optional decision-atlas example Use the atlas for the whole set, the map when only relationships matter, and the deliberation tree for one record. Do not generate a separate page when the built-in atlas already answers the question. The default `adrkit atlas` output uses [the decision-atlas example](examples/decision-atlas.html) through its `atlas:` slots. Open the example when the page needs adapting beyond what the slots fill. It is an offline reference snapshot of ADR Kit's own public records, not a template that must be followed. Inspect only the layout or interaction code needed; the embedded sample data is not instruction context. Reuse the useful parts; sections, chapter-bar entries, and legend or timeline keys whose data is absent hide themselves at runtime, the chapter bar renumbers, so no empty block survives. A timeline, relationship map, local-search directory, reference counts, and replacement chains are options, not a required checklist. Generate a fresh artifact from the target repository's records and the CLI's relationship output; replace all sample records, counts, dates, titles, paths, and links. Do not invent relationships, infer provenance, or copy another repository's contents. Keep archived history visually distinct from current authority, and publish archived bodies in full: the folder already carries the importance signal for agents, and a reader who opens an archived record wants its text. Resolve record links for the delivered file's location, as `atlas --map --format html --out` does. Preserve the `--record <name>` route; the example does not supply missing deliberations. The example marks its replaceable regions with `atlas:` comments — HTML around the title, hero, statistics, section notes, map, and footer, and a JavaScript comment around the records array. Replace what is inside a marker; keep the markers only when the page should stay regenerable from the same example. To build one from the example, follow these steps in order: 1. Read only `adr/implemented/` and `adr/archived/` records — the same scope as `adrkit atlas`. Build one records entry per file: the front matter fields in order, then `number` and `title` from the `# ADR:` line, the raw Markdown body (archived bodies in full), `path`, `source` (the record's blob URL at the pinned commit), and `lifecycle`. Determine `lifecycle` from the containing directory: `implemented` for `adr/implemented/`, `archived` for `adr/archived/`, regardless of `status`. Ordinary archives retain `status: implemented`; only records with `superseded-by` have a successor. 2. Run `adrkit atlas --map --format html` and keep its `<svg>` inline together with its canvas script. Rewrite each node link to an in-page `#adr-N` anchor and give the group the same id; the reader dialog, timeline, and directory search read the records array and the map edge attributes, so keep the class names and data attributes intact. 3. Compute the section notes from the repository's own numbers: record count, the first-to-last created date span and the number of distinct days that carry records, the densest day, the most-referenced implemented record, and the archived count. Never carry another repository's numbers into prose. The hero names the project once: the h1 is the name the repository declares for humans — its README or site title, else the repository name as written, in its Chinese form when the repository declares one; never translate, abbreviate, or substitute a suffixed package name. With the name in the h1, the eyebrow states only the genre and the lead paragraph opens with the archive's substance. 4. In the footer, pin the commit the records were read from (`git rev-parse --short HEAD`) and link it to the remote tree URL, so the page is traceable to an exact state. 5. Keep the example's legend, reader dialog, pagination, and back-to-top behavior and structure unchanged — they are repository-agnostic; translate their visible copy into the user's language, and vary the data and the short prose blocks. For an adapted long page, keep navigation compact and the directory search local. Vertical wheel and finger swipes should scroll the page; explicit drag or horizontal gestures pan the map, while pinch or Ctrl/⌘ + wheel zooms. Bound panning on both axes and center drawings that fit. Match the user's language and styling preferences. Keep one vertical rhythm: every section boundary should read as the same gap, so wrapper padding never stacks with section padding. Check the resulting page at desktop and mobile sizes, including record opening, navigation, search, canvas bounds, and page scrolling over the map; after sections hide, confirm the chapter bar renumbered contiguously and every entry still resolves; keep it usable offline. When the user requests HTML visualization, open the generated file in the default browser if available and return a clickable file path. If the user asks for a file only or says not to open it, leave the browser alone. If opening is unavailable or fails, explain that briefly and still return the file path. The CLI emits HTML to stdout unless `atlas --out` is used; the agent owns the browser handoff.
From luochang212/adr-kit, MIT.
More from luochang212/adr-kit
- 0.587score
adrkit-grill
luochang212
Use when a decision is still being shaped, or the user asks to grill a plan; interrogate the design against the active records until no branch is silently…
SkillAFreshness: A★ 1Claude Code +9 - 0.586score
adrkit-init
luochang212
Use when initializing ADR Kit in a repository or when the agent cannot find an adr/ directory.
SkillAFreshness: A★ 1Claude Code +9 - 0.582score
adrkit-review
luochang212
Use when checking whether changed code and changed decision records still agree; scoped to record coherence, not general code review.
SkillAFreshness: A★ 1Claude Code +9 - 0.582score
adrkit-note
luochang212
Use when you notice a choice future work must live with but cannot shape a record right now — park it as one line in the buffer and keep moving; the buffer is…
SkillAFreshness: A★ 1Claude Code +9 - 0.582score
adrkit-propose
luochang212
Use when you catch yourself settling a choice future work must live with — a dependency, format, boundary, or default — and adr/ does not already own it; the…
SkillAFreshness: A★ 1Claude Code +9 - 0.582score
adrkit-record
luochang212
Use when a choice that will constrain future work has already shipped and adr/ does not own it yet; record it directly, with no pending proposal.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
adrkit-implement
luochang212
Use when a proposal has shipped and should become a numbered, active decision.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
adrkit-reject
luochang212
Use when a formal proposal is declined and its anti-pattern should be recorded.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
adrkit-supersede
luochang212
Use when one implemented decision fully replaces another and the old record must become archived history.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
adrkit-validate
luochang212
Use when checking ADR record format before implementation, rejection, archival, or commit.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
adrkit-archive
luochang212
Use when an implemented record no longer guides work and current behavior has another authoritative owner.
SkillAFreshness: A★ 1Claude Code +9
Signals
- Stars
- 1
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Oct 23, 2026 (13 days ago)
- Last release
- Oct 23, 2026 (13 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/luochang212/adr-kit
- Website
- www.luochang.ink/adr-kit/
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from luochang212
- 0.587score
adrkit-grill
luochang212
Use when a decision is still being shaped, or the user asks to grill a plan; interrogate the design against the active records until no branch is silently…
SkillAFreshness: A★ 1Claude Code +9 - 0.586score
adrkit-init
luochang212
Use when initializing ADR Kit in a repository or when the agent cannot find an adr/ directory.
SkillAFreshness: A★ 1Claude Code +9 - 0.582score
adrkit-review
luochang212
Use when checking whether changed code and changed decision records still agree; scoped to record coherence, not general code review.
SkillAFreshness: A★ 1Claude Code +9 - 0.582score
adrkit-note
luochang212
Use when you notice a choice future work must live with but cannot shape a record right now — park it as one line in the buffer and keep moving; the buffer is…
SkillAFreshness: A★ 1Claude Code +9 - 0.582score
adrkit-propose
luochang212
Use when you catch yourself settling a choice future work must live with — a dependency, format, boundary, or default — and adr/ does not already own it; the…
SkillAFreshness: A★ 1Claude Code +9 - 0.582score
adrkit-record
luochang212
Use when a choice that will constrain future work has already shipped and adr/ does not own it yet; record it directly, with no pending proposal.
SkillAFreshness: A★ 1Claude Code +9
Done for you
Rather not set it up yourself?
Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.
See how we build AI agents →Read the case studies
Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.
- launches shipped
- 50+
- launches shipped
- media placements
- 500+
- media placements
- founders on every account
- 2
- founders on every account
Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block
Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.
Get adrkit-visualize running in your business
Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.
Cite this page
adrkit-visualize skill. Agent Mart. https://agentmart.dev/skills/luochang212-adrkit-visualize (last updated 2026-10-24)