本文へ移動
cccskills
無料GitHub で公開

wiki-maintainer

Build and maintain a persistent, interlinked markdown wiki (second brain / research notebook / personal knowledge base) from raw sources. Use this skill when the user wants to ingest an article / paper / note / URL / PDF into their wiki, ask a question against their wiki, lint/health-check their wiki, or set up a new wiki from a folder. Triggers include phrases like "add this to my wiki", "ingest this", "file this article", "update my notes with this", "what does my wiki say about X", "health-check my wiki", "find stale/orphan pages", or any workflow involving a `raw/` source folder, `wiki/` markdown folder, `index.md`, or `log.md`. Compatible with Obsidian vaults.

インストール方法を見る

含まれるファイル(2)

  • SKILL.md9.4 KB
  • agents/openai.yaml1.1 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Wiki Maintainer

Treat the workspace as a persistent, compounding knowledge base. The human curates sources and asks questions; you do all the reading, summarizing, cross-referencing, filing, and bookkeeping.

When the task is a single clear sub-operation, prefer the specialized skills:

  • $wiki-ingest for filing one new source
  • $wiki-query for answering from the wiki
  • $wiki-lint for maintenance and health checks
  • $wiki-bootstrap for initializing or restructuring the wiki

Use this umbrella skill when the task spans multiple operations or when you need the overall workflow contract.

Vault tool

Interpreter exposes the note graph through interpreter_vault.

Use it before large wiki operations:

  • action="snapshot" to confirm workspace scale and graph shape
  • action="search_notes" to find relevant notes by title, alias, or tag
  • action="note_context" to inspect backlinks, outgoing links, tags, and broken links for one note
  • action="notes_with_tag" and action="list_tags" for tag-driven navigation
  • action="lint" for structural wiki health checks

Do not manually rebuild the note graph with ad hoc grep when interpreter_vault already has the graph.

Architecture

Default greenfield layout:

  • raw/ — immutable source documents (clipped articles, PDFs, notes, transcripts). Read-only. Never write to raw/.
  • wiki/ — your output: summary, entity, concept, and synthesis pages. You own this layer entirely.
  • index.md and log.md at the workspace root — navigation and history.

If the workspace already has an established markdown/wiki structure, adopt it instead of forcing a parallel raw/ + wiki/ tree. Respect the user's existing folders, daily-note location, frontmatter conventions, and note naming. Only bootstrap the default layout when the workspace is actually empty or when the user explicitly asks for the raw/ + wiki/ structure.

Page Types (in wiki/)

Every wiki page is a markdown file with YAML frontmatter. Use hyphen-case filenames matching the title.

---
title: Page Title
type: source | entity | concept | synthesis | comparison | overview
created: YYYY-MM-DD
updated: YYYY-MM-DD
sources: [raw/path/to/source1.md, raw/path/to/source2.pdf]
tags: [tag1, tag2]
---
  • source — one per ingested document. Title = source title. Includes citation, 3–8 bullet takeaways, key quotes with page/section refs, open questions, and [[wikilinks]] to every entity/concept it touches.
  • entity — a person, organization, place, product, paper, event. One page per entity across all sources.
  • concept — an idea, mechanism, theory, method, pattern. One page per concept.
  • synthesis — a multi-source analysis, thesis, or evolving position. Cites sources with [[wikilinks]].
  • comparison — side-by-side table or contrast across two+ things.
  • overview — a top-level framing page for a topic area.

Link everything. Use [[Page Name]] wikilinks (Obsidian-compatible). If you reference something that doesn't have a page yet, link it anyway — orphan links surface what to create next.

Operations

Ingest

When the user drops a new source into raw/ (or gives you a URL) and asks to ingest it:

  1. Read the source fully. If it's a URL and a web-fetch tool is available, fetch it. If it's a PDF, extract text. If it's a markdown clip (Obsidian Web Clipper), read it directly. If images are referenced and the user asks, view them separately.
  2. Discuss key takeaways briefly with the user (3–6 bullets) before writing. Ask what to emphasize only if the angle is genuinely ambiguous.
  3. Create the source page at wiki/sources/<title-slug>.md with frontmatter, citation, takeaways, quotes, and [[wikilinks]].
  4. Touch every related page. For each entity, concept, or claim mentioned: create the page if missing, or update the existing page with the new information. Note contradictions explicitly with a > [!warning] Contradiction callout or inline ⚠️ note, citing both sources.
  5. Update index.md with a new entry under the appropriate section.
  6. Append to log.md with the standard prefix format.
  7. Report back to the user: which pages were created, which were updated, and any contradictions flagged. Be concise.

A durable wiki workflow must already exist before these steps begin. If not, bootstrap first with $wiki-bootstrap.

A single ingest typically touches 5–15 wiki pages. That is expected and correct.

Query

When the user asks a question of the wiki:

  1. Read index.md first to find relevant pages.
  2. Drill into those pages. Follow [[wikilinks]] as needed.
  3. Answer with citations. Every claim should point to a wiki page (which itself cites raw/ sources). Use the form ([[Page Name]]) inline.
  4. Offer to file the answer back as a new wiki page if it's a synthesis, comparison, or non-trivial analysis worth keeping. Exploration outputs should compound in the wiki, not disappear into chat history.

Lint

When the user asks for a health-check:

  1. Contradictions — scan for pages where claims conflict with newer sources.
  2. Stale claims — pages not updated since their sources were superseded.
  3. Orphans — pages with no inbound links.
  4. Dangling links — [[wikilinks]] pointing to pages that don't exist yet.
  5. Missing concepts — important terms mentioned repeatedly but lacking their own page.
  6. Gaps — areas where more sources would strengthen the synthesis.

Report findings as a prioritized list with suggested next actions. Do not auto-fix unless the user asks.

index.md Format

Content-oriented catalog. Organized by section. One line per page:

# Index

## Synthesis
- [[Evolving thesis on X]] — current position across 12 sources

## Overviews
- [[Topic A]] — framing page for topic A

## Entities
- [[Person Name]] — short description
- [[Org Name]] — short description

## Concepts
- [[Concept Name]] — one-line hook

## Sources
- [[Source Title]] — author, date, 1-line summary

Update on every ingest. Keep one-line summaries tight (under ~80 chars).

log.md Format

Chronological, append-only, grep-friendly. Every entry starts with the exact prefix:

## [YYYY-MM-DD] ingest | Source Title
- Created: [[Source Title]], [[New Entity]]
- Updated: [[Existing Concept]], [[Other Page]]
- Contradiction: [[Claim X]] (new source disagrees with [[Older Source]])

## [YYYY-MM-DD] query | "user's question verbatim or paraphrased"
- Filed as: [[Resulting Analysis Page]] (optional)

## [YYYY-MM-DD] lint | summary
- 3 orphans, 2 dangling links, 1 contradiction flagged

The prefix ## [YYYY-MM-DD] <op> | must be stable so grep "^## \[" log.md works.

Hard Rules

  • Never modify raw/. Sources are immutable. If a source is wrong, note it on the corresponding wiki/sources/ page.
  • Always update index.md and log.md on every ingest. No silent changes.
  • Always use [[wikilinks]], not markdown [text](path.md) links, for inter-wiki references. Obsidian compatibility matters.
  • Flag contradictions, don't resolve them silently. The human decides which side wins.
  • Cite sources. Every non-trivial claim on an entity/concept/synthesis page must trace back to a raw/ source via its wiki/sources/ page.
  • Don't re-summarize from scratch. Before writing, read the existing page (if any) and build on it. The wiki is a compounding artifact.
  • Prefer touching more pages over fewer. If a source mentions 8 entities, update all 8 pages.
  • Keep the human in the loop on sourcing. Do not auto-fetch and auto-ingest URLs the user did not hand you.
  • Respect the existing vault structure. Do not create a second note system beside the user's real one unless they asked for that.

First-Run Setup

If the workspace is empty or clearly greenfield and has no raw/, wiki/, index.md, or log.md:

  1. Create raw/ and wiki/ directories.
  2. Create index.md with empty section headers.
  3. Create log.md with a single entry: ## [YYYY-MM-DD] setup | wiki initialized.
  4. Tell the user the wiki is ready and suggest they drop their first source into raw/ (or clip one with Obsidian Web Clipper).

If the workspace already has a real markdown/wiki layout, skip bootstrap and work inside that layout.

Obsidian Compatibility

This wiki is designed to work inside an Obsidian vault. [[wikilinks]], YAML frontmatter, and the default raw/ + wiki/ layout all render natively, but when the user already has a working vault structure, respect that structure instead of replacing it. The user can browse the graph view, use backlinks, and edit pages directly — you maintain consistency from the other side.

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Build a period-end accrual schedule by computing each accrual, citing support, and drafting journal entries for controller approval. Use during month-end close; this drafts entries only and must not post them.

日本語の概要は準備中です。原文の説明を表示しています。

openinterpreter/interpreter-workstation222026年10月10日 更新

audit-xls

無料

Audit a spreadsheet for formula accuracy, errors, and common financial-model mistakes. Use for selected ranges, single sheets, or whole-workbook model checks including balance-sheet balance, cash tie-out, roll-forwards, and logic sanity.

日本語の概要は準備中です。原文の説明を表示しています。

openinterpreter/interpreter-workstation222026年10月10日 更新

Use this skill when the user needs advanced Playwright control of an already running browser session through Interpreter's app-managed browser bridge and the `builtin-js-repl` `js_repl` tool, after simple browser page work cannot be handled by the unified `builtin-interpreter` browser page tools.

日本語の概要は準備中です。原文の説明を表示しています。

openinterpreter/interpreter-workstation222026年10月10日 更新

Drive a native desktop app through Interpreter's builtin-cua-driver. Use get_app_state, click/type/scroll/drag, and verify by calling get_app_state again when the user asks to operate a real desktop app, browser chrome, native dialog, menu, secure prompt, file chooser, or hidden/background window.

日本語の概要は準備中です。原文の説明を表示しています。

openinterpreter/interpreter-workstation222026年10月10日 更新

doc

無料

Create, inspect, and edit Word documents (`.docx`) with local code-execution tools. Use for document drafting, structured edits, tables, comments, and layout-sensitive Word deliverables.

日本語の概要は準備中です。原文の説明を表示しています。

openinterpreter/interpreter-workstation222026年10月10日 更新

Design or revise Interpreter UI for non-developers with an ultra-minimal, utilitarian style inspired by the OpenAI website and ChatGPT, not the developer platform. Use when simplifying settings, onboarding, chat, forms, empty states, or helper copy; when removing extra containers, labels, or visual noise; when replacing technical wording with plain language; when clarifying hierarchy and making the main user action obvious; when making every visible label and status understandable to an average HR admin or other non-technical user; when hiding implementation details such as logs, IDs, endpoints, paths, or protocol names; or when a screen must respect the app's existing CSS variables and light/dark theme behavior.

日本語の概要は準備中です。原文の説明を表示しています。

openinterpreter/interpreter-workstation222026年10月10日 更新

openinterpreter のスキルをすべて見る

このスキルの問題を報告する