Agent skill

Docs Lead

by lablup in lablup/backend.ai-webui

A skill your agent uses whenever the user mentions docs, the manual, documentation, terminology, translations, or screenshots — including indirect mentions like "이 PR 문서 영향 봐줘", "문서 점검", "용어 통일"…

LGPL-3.0Auto-check passedWriting & Content

Install Docs Lead

skills CLI
$ npx skills add lablup/backend.ai-webui --skill docs-lead -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install lablup/backend.ai-webui docs-lead --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/lablup/backend.ai-webui.git skills-src && mkdir -p .claude/skills && cp -r skills-src/.claude/skills/docs-lead .claude/skills/docs-lead && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
docs-lead
GitHub stars
133
Token cost
~2.9k tokens
SKILL.md length
1,260 words
Files
5 (incl. references)
Skills in repo
13
Repo updated
First seen
Licence
LGPL-3.0

At a glance

A skill your agent uses whenever the user mentions docs, the manual, documentation, terminology, translations, or screenshots — including indirect mentions like "이 PR 문서 영향 봐줘", "문서 점검", "용어 통일"…

  • Works in 6 steps: Load prior state → Fresh health diagnosis → Build the work queue → …
  • The user mentions docs
  • SKILL.md covers Reference files, Domain context, Karpathy mapping (one-line) and Workflow, plus 2 more sections
  • Calls git, pnpm and gh

What it does

Docs Lead is an agent skill from lablup/backend.ai-webui. Use whenever the user mentions docs, the manual, documentation, terminology, translations, or screenshots — including indirect mentions like "이 PR 문서 영향 봐줘", "문서 점검", "용어 통일", "/docs-lead", or any phrasing about updating the Backend.AI WebUI user manual under packages/backend.ai-webui-docs/. Single entry point that runs docs-lint diagnosis, surfaces a prioritized queue via AskUserQuestion, and orchestrates the four docs worker subagents (planner / writer / screenshot-capturer / reviewer) through approval gates…

Its SKILL.md is about 2.9k tokens, which your agent loads only when the skill is triggered. The skill folder holds 5 other files, including reference files (for example `references/anti-patterns.md`, `references/decision-gate-templates.md` and `references/karpathy-mapping.md`).

It sits in Writing & Content, covering Translation, Technical writing and Linting and formatting. The repository describes itself as: Backend.AI Web UI for web / desktop app (Windows/Linux/macOS). Backend.AI Web UI provides a convenient environment for users, while allowing various commands to be executed… The licence is LGPL-3.0.

When your agent uses it

  • The user mentions docs
  • Screenshots — including indirect mentions like 이 PR 문서 영향 봐줘
  • Any phrasing about updating the Backend.AI WebUI user manual under packages/backend.ai-webui-docs/

Example prompts

  • “이 PR 문서 영향 봐줘”
  • “/docs-lead”
  • “/docs-lead”

Workflow steps

6 steps, taken from the step headings in SKILL.md.

  1. Load prior state
  2. Fresh health diagnosis
  3. Build the work queue
  4. Decision gate
  5. Orchestrate the chosen item
  6. Persist state, decide what's next

What it can do on your machine

Read from SKILL.md and the folder at commit 10554dc. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Shell commands in SKILL.md call:

    • git
    • pnpm
    • gh

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md. Its commands use git, pnpm and gh, which can reach the network depending on how they are called.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Docs Lead loads about 2.9k tokens when it runs, and up to ~7.4k if it reads all its reference files. Until then it costs about 161 tokens; SKILL.md has 1,260 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~161
When it runs · the whole SKILL.md, loaded when a task matches
~2.9k
With references · SKILL.md plus every file in references/, read only if the agent opens them
~7.4k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from lablup/backend.ai-webui at commit 10554dc, republished under its LGPL-3.0 licence (© lablup). 1,260 words, ~2,866 tokens.

Download SKILL.mdSave it as .claude/skills/docs-lead/SKILL.md (or your agent's skills folder). This skill also uses 4 other files; get the full folder from GitHub.
name
docs-lead
description
Use whenever the user mentions docs, the manual, documentation, terminology, translations, or screenshots — including indirect mentions like "이 PR 문서 영향 봐줘", "문서 점검", "용어 통일", "/docs-lead", or any phrasing about updating the Backend.AI WebUI user manual under packages/backend.ai-webui-docs/. Single entry point that runs docs-lint diagnosis, surfaces a prioritized queue via AskUserQuestion, and orchestrates the four docs worker subagents (planner / writer / screenshot-capturer / reviewer) through approval gates. Reading the manual to answer a question is the `bai-agent` skill's job; this skill is for writing and maintaining it.

docs-lead — Documentation Team Lead

Category note: This is a skill (runs in the main context with full access to AskUserQuestion and the Agent tool), not a subagent. The "Team Lead" metaphor describes the role, not the technical type — do not look for this under .claude/agents/. The skill lives at .claude/skills/docs-lead/.

You are the docs team lead. The user makes one request and you handle the docs lifecycle — diagnosis, prioritization, planning, writing, screenshots, review — with explicit decision gates at every irreversible step. You are the only caller of the docs worker agents. Users should not invoke the workers directly while this skill is active. If they did (rare), take over: read the worker's output file, restore orchestration, and continue.

Reference files

Read these on demand — they hold the bulk of the patterns so this SKILL.md stays scannable.

  • references/karpathy-mapping.md — three-layer/three-op model and why Query is excluded. Read when reasoning about design or extension.
  • references/worker-invocation-templates.md — copy-ready Agent(...) prompts for the four workers. Read in step 5 of the workflow.
  • references/decision-gate-templates.md — AskUserQuestion shape rules and examples for the four gates. Read in step 4 / 5a / 5c / 6.
  • references/anti-patterns.md — shortcuts that corrupt the workflow over time. Read when you're tempted to skip a gate or chain topics.

Domain context

  • Documentation root: packages/backend.ai-webui-docs/src/{en,ko,ja,th}/
  • Schema files (the editorial contract):
    • packages/backend.ai-webui-docs/TERMINOLOGY.md
    • packages/backend.ai-webui-docs/DOCUMENTATION-STYLE-GUIDE.md
    • packages/backend.ai-webui-docs/TRANSLATION-GUIDE.md
    • packages/backend.ai-webui-docs/SCREENSHOT-GUIDELINES.md
    • packages/backend.ai-webui-docs/src/book.config.yaml
  • State files (you own these): packages/backend.ai-webui-docs/.agent-output/
    • docs-state.md — append-only work log (rotates at 30 blocks)
    • docs-lint-report.md — last 10 lint runs (rewrite-with-rotation, 256 KB cap)
    • docs-update-plan-{topic}.md, docs-review-report-{topic}.md — per-topic worker outputs
    • archive/ — rotated entries

Why project-local state, not user-level ~/.claude/plugins/data/...? docs state is project-and-worktree-bound: the queue references PR numbers, file paths, and topic slugs that only make sense in this repo. It must live alongside the existing docs-update-plan-{topic}.md / docs-review-report-{topic}.md files the workers already produce, so hand-offs are coherent.

Karpathy mapping (one-line)

Partial adoption of the LLM Wiki pattern (gist 442a6bf555914893e9891c11519de94f): Ingest and Lint ops adopted, Query intentionally excluded. See references/karpathy-mapping.md for the full mapping and rationale.

Workflow

Execute in order. Do not skip steps.

Step 1 — Load prior state
bash
cat packages/backend.ai-webui-docs/.agent-output/docs-state.md 2>/dev/null || echo "(no prior state)"

Read what was last worked on, what is in flight, what was deferred. Missing file = empty queue.

Step 2 — Fresh health diagnosis

Reuse rule first: if docs-lint-report.md exists with a top-section ISO timestamp within the last 10 minutes AND git status --short -- packages/backend.ai-webui-docs/src/ is clean, skip the lint pass and reuse the existing report. Lint reads dozens of files; don't burn that cost twice in a row.

Otherwise, invoke docs-lint via the Agent tool with subagent_type: "docs-lint". Prompt: "Run a full lint pass and update packages/backend.ai-webui-docs/.agent-output/docs-lint-report.md per your rotation policy. Then summarize the top of the new run section (counts only) in your response."

Then read just the top run section:

bash
sed -n '/^---RUN---/,/^---RUN---/p' packages/backend.ai-webui-docs/.agent-output/docs-lint-report.md | head -120
Step 3 — Build the work queue

Merge three sources into one prioritized list:

  1. User-explicit ask (highest priority) — specific PR, feature, or area the user pointed at.
  2. Lint findings — group by check type. PR coverage gaps are usually highest-leverage; terminology drift is lowest-effort.
  3. In-flight from state — items marked status: in-progress or status: deferred.

Attach a one-line scope estimate to each item: which files change, which worker chain runs, whether screenshots are needed (live app required), which languages.

Step 4 — Decision gate

Present 2–4 prioritized options via AskUserQuestion. Shape rules and examples live in references/decision-gate-templates.md (Gate 1). Read that file when constructing this gate.

Special cases:

  • Queue empty → do not call AskUserQuestion. Report "no work needed — most recent lint on <timestamp> is clean" and stop.
  • More than 4 candidates → bundle lower-priority into "기타 N건 일괄 처리".
  • More than 5 lint findings → include a "전체 lint 리포트 먼저 보여줘" option so the user can self-triage.
Step 5 — Orchestrate the chosen item

The chain: planner → [user confirm] → writer → [user confirm if screenshots] → screenshot-capturer → reviewer. Reviewer runs automatically after writer; every other transition needs an explicit AskUserQuestion confirmation.

For the actual Agent(...) invocation prompts for each worker, see references/worker-invocation-templates.md. Each prompt must be self-contained (workers don't see your context) and use a consistent kebab-case topic slug across the whole chain.

5a. Planner → plan confirmation gate
  1. Invoke docs-update-planner (see template).
  2. Read the resulting packages/backend.ai-webui-docs/.agent-output/docs-update-plan-{topic}.md.
  3. Summarize to the user in ≤4 bullets: files that will change, new sections, screenshots needed, languages targeted.
  4. Decision gate — see references/decision-gate-templates.md Gate 2.
5b. Writer

Only after the user confirms 5a. Invoke docs-update-writer (see template). After it returns, run git status --short -- packages/backend.ai-webui-docs/src/ yourself to see what actually changed — do not trust the worker's claimed file list.

5c. Screenshot capture — live-app gate

Only if the plan calls for screenshots. Decision gate first — see references/decision-gate-templates.md Gate 3. If user picks "지금 캡처", invoke docs-screenshot-capturer (see template). Skip this entire substep if the plan listed zero screenshots and writer left no TODO markers.

Show full SKILL.md (484 more words)Show less
5d. Reviewer (automatic)

Invoke docs-update-reviewer (see template). Summarize the report — critical / warning / minor counts. If any "critical" exists, surface it to the user before declaring the topic complete.

Step 6 — Persist state, decide what's next

Append a single block to packages/backend.ai-webui-docs/.agent-output/docs-state.md:

markdown
## <ISO-8601 timestamp> — <topic slug>

- Source: <PR # / feature description / lint-driven>
- Status: completed | partial | deferred
- Plan: packages/backend.ai-webui-docs/.agent-output/docs-update-plan-<topic>.md
- Review: packages/backend.ai-webui-docs/.agent-output/docs-review-report-<topic>.md
- Worker chain: planner → writer → [screenshot-capturer] → reviewer
- Files changed: <count> across <languages>
- Outstanding: <one-line if partial; omit otherwise>

Rotation: if docs-state.md now has more than 30 ## blocks, move the oldest blocks into packages/backend.ai-webui-docs/.agent-output/archive/docs-state-YYYY-QN.md (calendar quarter of the moved entries' first timestamp; append to the archive if it exists).

Then decide what's next — see references/decision-gate-templates.md Gate 4. If queue is empty, do not gate; just report and stop.

Hard rules

  • Approval gate before writing. Never call docs-update-writer without an explicit AskUserQuestion confirmation immediately preceding it. Lint findings alone are not consent.
  • Live-app gate before screenshots. Never call docs-screenshot-capturer without an explicit live-app readiness confirmation.
  • Workers never call other workers. Each worker invocation is independent; hand-off goes through this skill and packages/backend.ai-webui-docs/.agent-output/.
  • No silent edits. This skill itself must not Edit docs files. Mutation is delegated to docs-update-writer. The only files this skill writes directly are packages/backend.ai-webui-docs/.agent-output/docs-state.md (and its archive/ rotations).
  • Idempotency. Re-invoking with no new changes produces no new queue items beyond what is already in state.
  • One topic per chain. Always return to the decision gate (step 4 or step 6) between topics. Multi-topic chains make rollback impossible and inflate context catastrophically.
  • Match the user's language. Reply in the language the user is using — Korean if they speak Korean, English if they switch. Worker outputs (plan files, lint reports, PR bodies) stay in English regardless; that is a documentation convention, not a conversation rule.

Gotchas

Real failure modes you will hit if you don't think about them up front.

  • docs-lint's i18n-key-drift signal is heuristic — it produces candidates, not confirmed stales. Never auto-recapture based on it alone. Surface to user, let them eyeball before triggering docs-screenshot-capturer.

  • docs-update-planner overwrites docs-update-plan-{topic}.md if you call it twice with the same topic slug. Before invoking, check:

    bash
    ls packages/backend.ai-webui-docs/.agent-output/docs-update-plan-<topic>.md 2>/dev/null

    If a plan exists, ask the user: continue from it, overwrite, or pick a different slug (e.g., suffix with -v2).

  • docs-screenshot-capturer requires the dev server live — get its real URL from the webui-connection-info skill (Portless picks the host and may not be on 1355). If pnpm dev isn't running, the worker fails partway and leaves orphan files in .playwright-mcp/. Always confirm liveness in Gate 3 and remind the user how to start the server if they're unsure.

  • git log --merges is empty in this repo because PRs are squash-merged. docs-lint already uses gh pr list for coverage gaps; if you ever extend coverage detection yourself, use the same path — not git log --merges.

  • AskUserQuestion always offers "Other". The user can type a free-text response even when your options don't cover their case. If they pick "Other" with a custom intent (e.g., "PR #7430 먼저 처리해줘"), treat that as a queue override and rebuild from step 3 — don't try to fit it into the closest existing option.

© lablup, LGPL-3.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 4 other files (references) in .claude/skills/docs-lead of lablup/backend.ai-webui.

  • SKILL.md
  • references/anti-patterns.md
  • references/decision-gate-templates.md
  • references/karpathy-mapping.md
  • references/worker-invocation-templates.md

Open the folder on GitHubat commit 10554dc

Compare with similar skills

Docs Lead next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Docs Lead compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Docs Lead this skilllablup/backend.ai-webui133—~2.9kAutomated safety check: PassLGPL-3.0
Translateforthecraft/drf-auth-kit121—~2.4kAutomated safety check: NotesMIT
Whitepaper Bare Section Prose FillerFlorianBruniaux/claude-code-ultimate-guide6.1k—~1.5kAutomated safety check: PassCC-BY-SA-4.0
Write Motrixlab Task DocsMotphys/MotrixLab152—~1.8kAutomated safety check: PassApache-2.0
Academic Prose De-AI Editorheise3/academic-deai209—~1.4kAutomated safety check: PassMIT
Aholo Viewer Docsmanycoretech/aholo-viewer1.1k—~341Automated safety check: PassMIT

Similar skills

  • Translate

    forthecraft/drf-auth-kit

    Run Django translation workflow - extract untranslated strings, translate them to 59 languages using parallel translator subagents, and apply back to PO files.

    121 GitHub stars~2.4k tokensUpdated 1 mo ago
    Writing & ContentAuto-check: notes
  • Whitepaper Bare Section Prose Filler

    FlorianBruniaux/claude-code-ultimate-guide

    Adds two to four sentences of intro prose before each bare heading in the French and English whitepapers, splitting the work across nine parallel agents.

    6.1k GitHub stars~1.5k tokensUpdated yesterday
    Writing & ContentAuto-check passed
  • Write Motrixlab Task Docs

    Motphys/MotrixLab

    Write, restructure, or review bilingual Sphinx/MyST user documentation for MotrixLab — task environments (from simple single-page tasks to reusable task families) and framework-level tutorial pages…

    152 GitHub stars~1.8k tokensUpdated today
    Writing & ContentAuto-check passed
  • Academic Prose De-AI Editor

    heise3/academic-deai

    Edits Chinese or English scholarly writing for natural phrasing and removes AI-sounding templated structure, while keeping claims, citations and author voice intact.

    209 GitHub stars~1.4k tokensUpdated 3 days ago
    Writing & ContentAuto-check passed
  • Aholo Viewer Docs

    manycoretech/aholo-viewer

    Guides writing and maintaining Aholo Viewer documentation: README, AGENTS.md, architecture notes, bilingual manual pages and AI collaboration guides.

    1.1k GitHub stars~341 tokensUpdated 13 days ago
    Writing & ContentAuto-check passed
  • Benchmark Translate

    shapeshift/web

    Run a quality benchmark of the /translate skill by selecting stratified test keys, capturing ground truth, translating, judging with sub-agents, and compiling a regression report.

    206 GitHub stars~1.6k tokensUpdated today
    Writing & ContentAuto-check passed

More from lablup/backend.ai-webui

All 13 skills in this repo
  • Walkthrough

    lablup/backend.ai-webui

    Mint a walkthrough for the PR this session just implemented: a set of numbered stops a reviewer opens in the live dev server, each one marking an element on screen with what changed and what to check.

    133 GitHub stars~4.9k tokensUpdated today
    Auto-check: notes
  • Backend AI Guide

    lablup/backend.ai-webui

    Expert guide for Backend.AI distributed computing platform. An agent skill from lablup/backend.ai-webui.

    133 GitHub starsUsed in 1 repo~1.8k tokens
    Auto-check passed
  • Dev Server

    lablup/backend.ai-webui

    Start the project's development server (pnpm dev for backend.ai-webui; discovered from README/package.json elsewhere), deriving the header color, app name, default backend endpoint and login…

    133 GitHub stars~6.9k tokensUpdated today
    Auto-check: notes
  • Record E2E Gif

    lablup/backend.ai-webui

    Record Playwright e2e tests as one GIF per test case (video → ffmpeg palette GIF) and return a markdown table for a PR description.

    133 GitHub stars~907 tokensUpdated today
    Auto-check: notes
  • Relay Mutation Store Updates

    lablup/backend.ai-webui

    A skill your agent uses when writing if (success) updateFetchKey(), an onRequestClose handler, or any refetch after a mutation; when a setting modal handles both create and update behind one…

    133 GitHub stars~2.8k tokensUpdated today
    Auto-check passed
  • Release Train Prep

    lablup/backend.ai-webui

    Post a Korean release risk digest to a Microsoft Teams thread, grouped by risk category.

    133 GitHub stars~5k tokensUpdated today
    Auto-check passed

Questions about Docs Lead

What does Docs Lead do?

A skill your agent uses whenever the user mentions docs, the manual, documentation, terminology, translations, or screenshots — including indirect mentions like "이 PR 문서 영향 봐줘", "문서 점검", "용어 통일"…. ai-webui.ai-webui-docs/.

When should I use Docs Lead?

Docs Lead fits situations like: the user mentions docs; screenshots — including indirect mentions like 이 PR 문서 영향 봐줘; any phrasing about updating the Backend.AI WebUI user manual under packages/backend.ai-webui-docs/.

How do I install Docs Lead in Claude Code?

Run `npx skills add lablup/backend.ai-webui --skill docs-lead -a claude-code`. Or copy the skill folder (.claude/skills/docs-lead in lablup/backend.ai-webui) into .claude/skills/docs-lead in your project. Claude Code loads it when a task matches its description.

How do I install Docs Lead in Codex?

Run `npx skills add lablup/backend.ai-webui --skill docs-lead -a codex`. Or copy the skill folder (.claude/skills/docs-lead in lablup/backend.ai-webui) into .agents/skills/docs-lead in your project. Codex loads it when a task matches its description.

Can I use Docs Lead in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add lablup/backend.ai-webui --skill docs-lead -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/docs-lead, .gemini/skills/docs-lead, .github/skills/docs-lead and .opencode/skills/docs-lead in your project.

What does Docs Lead need to run?

Going by SKILL.md and its folder, Docs Lead needs the command-line tools its instructions call (git, pnpm and gh).

Does Docs Lead access the network?

SKILL.md contains no URLs. Its commands use git and gh, which can reach the network depending on how they are called. This is read from the text; nothing was executed.

Is Docs Lead safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Docs Lead use?

Docs Lead is published under the LGPL-3.0 licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Docs Lead use?

About 2.9k tokens (SKILL.md is roughly 11k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full. Its references folder adds about 4.6k tokens, read only when the agent opens those files.

What are the alternatives to Docs Lead?

Skills that share tags, products or a category with Docs Lead: Translate (forthecraft/drf-auth-kit, 121 stars), Whitepaper Bare Section Prose Filler (FlorianBruniaux/claude-code-ultimate-guide, 6.1k stars), Write Motrixlab Task Docs (Motphys/MotrixLab, 152 stars) and Academic Prose De-AI Editor (heise3/academic-deai, 209 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Docs Lead?

lablup (a GitHub organization) maintains it in lablup/backend.ai-webui, which has 133 GitHub stars. The repository holds 13 skills in this directory. The repository was last updated on October 7, 2026.

Source: lablup/backend.ai-webui on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.