본문으로 건너뛰기
cccskills
무료GitHub에 공개

update-project-docs

MANDATORY for every coding agent (Claude Code, Codex, or any other) — keep this repository's documentation in sync after any change to behavior, configuration, interfaces, events, schema, or features. Use automatically at the end of change-sets that alter an env var, event type, hook or state behavior, API, schema, WebSocket message, MCP tool, CLI command, or user-facing feature, and whenever the user asks to update documentation.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

설치 방법 보기

포함된 파일(4)

  • SKILL.md10.2 KB
  • references/doc-map.md6.9 KB
  • scripts/doc-coverage.sh1.8 KB
  • scripts/wiki-block-lengths.sh3.4 KB

SKILL.md(원문)

설치하기 전에 에이전트에게 전달되는 지시 내용을 확인할 수 있습니다.

Update Project Docs

This repository keeps an unusually large, multi-surface, multi-language doc set. Docs drift silently because a change often belongs in 6–10 files across 5 languages plus two HTML pages. This skill encodes which docs exist, which change-types touch which docs, and how to propagate consistently (including the wiki i18n + cache-bump dance).

Authoritative inventory with exact section anchors lives in references/doc-map.md — read it when deciding where a specific change lands. The repo rules .claude/rules/docs-markdown.md ("update all affected docs together"), .claude/rules/wiki-i18n.md, and .claude/rules/i18n-parity.md are binding. Translation propagation — the mirrored READMEs, the wiki bundles, and the UI keys — is owned by the i18n-parity skill; run it whenever this skill's mapping sends you into a localized file.

When to update (including without being asked)

Update docs in the same change-set (PR/commit) as the code, before claiming done — do not wait for the user to ask — whenever the change is observable from outside the module:

  • New/changed env var → every env-var table + .env.example.
  • New event type (e.g. an events.event_type value) → every event-type list/table.
  • New/changed hook behavior or session/agent state transition → hook docs + every state-machine diagram.
  • New/changed API route or response shape → API docs + route tables + OpenAPI.
  • DB schema change (table/column/index) → database docs + ERD.
  • New WebSocket message type → client/server WS docs.
  • New MCP tool → MCP docs.
  • New CLI command / script / renamed file referenced in docs → command lists + onboarding guides.
  • New user-facing feature / page / background service → feature tables + landing + wiki + architecture.

Do NOT auto-update for: pure internal refactors with no observable/interface/config change, test-only changes, comment/typo fixes, or work the user explicitly scoped as "no docs". When unsure whether a change is observable, check the mapping below; if it touches any row, update.

Change → docs mapping

Change typeDocs to update
Env varREADME-EN.md, README-CN.md, README-VN.md, README-KO.md, README-ES.md (env tables), ARCHITECTURE.md (inline), server/README.md, wiki/index.html (env table) + wiki i18n, .env.example
Event typeREADME-EN.md+CN+VN+KO+ES (hook-event table), ARCHITECTURE.md (Event types line), docs/PLUGINS.md, wiki/index.html + i18n, docs/DATABASE.md (if it enumerates types)
Hook behavior / state transitiondocs/HOOKS.md, state-machine mermaid diagrams in README-EN.md+CN+VN+KO+ES + server/README.md + docs/DATABASE.md + wiki/index.html, ARCHITECTURE.md (hooks.js row)
API route / responsedocs/API.md, server/README.md (routes), ARCHITECTURE.md (routes row), server/openapi*.js (code)
DB schemadocs/DATABASE.md, ARCHITECTURE.md (ERD/schema)
WebSocket messageclient/README.md (Event Types), server/README.md, wiki/index.html
MCP toolmcp/README.md, docs/MCP.md
Feature / page / background serviceREADME-EN.md+CN+VN+KO+ES (feature table + data-flow list), concise README.md only when the capability is headline-level, ARCHITECTURE.md (module table), index.html (landing blurb), wiki/index.html + i18n, server/README.md or client/README.md
CLI command / scriptREADME-EN.md commands, concise README.md only for primary entry points, CLAUDE.md / AGENTS.md, INSTALL.md / SETUP.md
New languageRun the i18n-parity skill and work through its new-language checklist — it covers docs/I18N.md, client/src/i18n/**, the switchers, format.ts, README-<XX>.md, and the full wiki bundle

Procedure

  1. Classify the change against the table above. A change can hit multiple rows (a new feature with a new env var hits both).
  2. Write the canonical full English version first — usually README-EN.md and/or ARCHITECTURE.md. Get the wording right there; it anchors every full translation. Keep README.md as a concise English-only landing page.
  3. Propagate to translations README-CN.md, README-VN.md, README-KO.md, and README-ES.md: mirror the SAME edits at the corresponding sections. Keep identifiers, env-var names, event names, and code in English; translate only prose. Render "Waiting" as 等待中 (zh) / Đang chờ (vi) / 대기 중 (ko) / En espera (es). Match each file's existing terminology — read the neighboring lines first. The i18n-parity skill owns this propagation and its glossary.
  4. Landing page index.html: one concise marketing sentence in the most relevant existing feature card — light touch, no new sections.
  5. Wiki wiki/index.html: add the detailed prose/table/diagram at the length and in the position its neighbours already use (see Match the wiki's existing shape below), then follow .claude/rules/wiki-i18n.md — add zh + vi + ko + es entries for every new English string to wiki/i18n-content.js, then bump the cache: increment CACHE_NAME in wiki/sw.js and the i18n-content.js?v= query string in wiki/index.html. Skipping the cache bump means returning visitors never see the update.
  6. Area READMEs / docs/: update server/README.md, client/README.md, and the relevant docs/*.md per the mapping.
  7. Diagrams: when a state transition changes, edit every mermaid stateDiagram-v2 block that models it (they are duplicated across README/CN/VN/KO/ES, server/README, docs/DATABASE, wiki). Keep transition labels consistent.

Match the wiki's existing shape

The wiki is a designed page, not a changelog: several of its blocks live in fixed-size boxes. A block written at 2–3x the length of its neighbours breaks the layout, so measure before you write, and copy the pattern you find.

  • Feature carousel cards (#feature-carousel .feature-card) share one fixed-height box. All 39 of them are a single <p>; the group runs 209–628 characters with a ~494 median, so write to roughly 450–550 characters (~65–80 words) — one paragraph, no lists, no sub-headings, no exhaustive enumeration of every keybinding and edge case. Say what the feature is and the two or three things that make it distinctive; the depth belongs in the feature's own section further down the page.
  • Screenshot captions (.screenshot-caption) run ~150–300 characters: the emoji, the bolded screen name, an em dash, one dense sentence.
  • Card order is editorial, not chronological. A new feature does NOT go first. Insert it where it belongs by importance among the existing cards — a newly shipped convenience feature belongs in the middle or later half of the carousel, not ahead of the dashboard, board, and session cards.
  • Deep, unabridged prose (full keyboard maps, degradation behaviour, group-by- group breakdowns) belongs in a normal <section> with h3 + <ul>, never crammed into a card or caption.
  • Run .claude/skills/update-project-docs/scripts/wiki-block-lengths.sh and confirm your block sits inside the group's budget before finishing.
  • Changing wiki CSS? Verify it in a browser that ran script.js (see .claude/rules/wiki-i18n.md) — the scroll-reveal pass adds classes at runtime, so a :not([class]) selector that looks right in the file and in jsdom can apply to nothing on the live page.
  • The same rule of thumb applies everywhere on the page: before adding a block of any kind, read the two blocks around it and match their length, tone, markup, and heading depth. Do not invent a new pattern for one entry.

Verify (do not skip)

  • Coverage: run scripts/doc-coverage.sh <new-term> [...] (e.g. the new env var / event type / identifier) and confirm every doc the mapping flags shows a HIT. The matrix is advisory — not every term belongs in every file — but a flagged doc reading 0 is a miss to fix.
  • Tables: markdown tables stay pipe-balanced (header column count == every row).
  • Mermaid: each edited block still parses (valid source --> target: label).
  • i18n: every new wiki English string resolves to zh, vi, ko, and es; cache versions bumped — CACHE_NAME in wiki/sw.js plus the ?v= query strings, which must match between wiki/index.html and the service-worker PRECACHE list (it matches on the full URL, query included, so a stale entry is simply never served).
  • Wiki block sizing: .claude/skills/update-project-docs/scripts/wiki-block-lengths.sh exits 0 — no carousel card or caption is an outlier, and any new card is placed by importance rather than dropped at the front.
  • Format/tests: run npm run format (or prettier --check on touched files); for any code touched, run the verification from CLAUDE.md (npm run test:server / test:client / mcp:typecheck).
  • State exactly which docs were updated and which were intentionally skipped (with reason), mirroring the repo's verification policy.

Tips

  • The fastest way to find where something already lives: grep -n "<existing-neighbor-term>" <doc> (e.g. grep an adjacent env var to find the env table). references/doc-map.md lists the stable anchors per file.
  • Parallelize translations + HTML across subagents when the change is large, but write the canonical English edit yourself first so the translations have a faithful source.
  • One language/area per subagent keeps edits reviewable and tables un-corrupted.

리뷰

아직 리뷰가 없습니다. 사용해 본 소감을 남겨 주세요.

댓글

로그인 없이 쓸 수 있습니다. 사용 후기, 질문, 작동하지 않은 환경 등을 편하게 남겨 주세요.

아직 댓글이 없습니다. 첫 댓글을 남겨 보세요.

보내면 바로 공개됩니다. 이메일 주소나 전화번호 같은 개인 정보는 쓰지 마세요. URL은 2개까지입니다.

0 / 1,000자

이 양식은 reCAPTCHA로 보호되며 Google 개인정보처리방침 및 서비스 약관이 적용됩니다.

같은 저장소의 스킬

Inspect fired CCAM alerts and manage alert rules for token thresholds, event patterns, inactivity, and status duration. Use when acknowledging alerts, creating or editing a rule, checking cooldowns, or connecting alert rules to webhook targets.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

hoangsonww/Claude-Code-Agent-Monitor1.1천2026년 10월 22일 업데이트

Identify anomalous sessions using Agent Monitor data — cost outliers from the pricing engine, token anomalies (cache miss spikes, compaction baseline surges), unusual event type ratios (PreToolUse/PostToolUse gaps, APIError clusters), behavioral deviations from workflow intelligence (complexity score outliers, error propagation anomalies), and sessions with abnormal metadata (extreme turn_count, high thinking_blocks, zero turn_duration).

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

hoangsonww/Claude-Code-Agent-Monitor1.1천2026년 10월 22일 업데이트

Produce a detailed report on APIError events from Agent Monitor data — counts over time, which sessions and models are affected, and the likely root cause (rate limits, overload/529, or context-window pressure) inferred from each event's summary and data payload. Use when API errors spike or when you need to explain why requests are failing.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

hoangsonww/Claude-Code-Agent-Monitor1.1천2026년 10월 22일 업데이트

benchmark

무료

Benchmark one session (or a small recent set) against the rolling average using Agent Monitor data — cost, total tokens, tool count, and workflow complexity score — and report where each metric lands as a percentile of the population. Tells you whether a session was normal, cheap, or an outlier. Use when judging whether a session was typical or out of band.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

hoangsonww/Claude-Code-Agent-Monitor1.1천2026년 10월 22일 업데이트

Define a spend budget for Claude Code and, optionally, create a cost alert rule that fires when usage crosses the limit, via POST /api/alerts/rules on the Agent Monitor dashboard. Reads current spend from /api/pricing/cost to size the budget sensibly and explains every rule field before writing. Use when setting a spend limit or wiring up a budget guardrail.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

hoangsonww/Claude-Code-Agent-Monitor1.1천2026년 10월 22일 업데이트

Analyze prompt-cache effectiveness for Claude Code usage from the Agent Monitor dashboard — cache hit rate (total_cache_read / (total_cache_read + total_input)), cache_write vs cache_read reuse, cache-read vs cache-write spend, and the sessions with the poorest reuse. Pulls token totals from /api/analytics, per-session detail from /api/sessions, and dollar splits from /api/pricing/cost. Use when diagnosing cache spend or deciding whether prompt caching is paying off.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

hoangsonww/Claude-Code-Agent-Monitor1.1천2026년 10월 22일 업데이트

hoangsonww의 스킬 모두 보기

이 스킬의 문제 신고