mooqii/OpenPanels / agent-resources/preset-skills/wiki-default
wiki-default
VerifiedUse when creating, adding to, editing, or maintaining a persistent, structured wiki from curated documents with the default Wiki method.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 2.8 KB |
| references/ingest-markdown-into-wiki.md | markdown | 3.5 KB |
| references/maintain-wiki.md | markdown | 2.5 KB |
| references/wiki-conventions.md | markdown | 8.9 KB |
references/wiki-conventions.md
View on GitHub# LLM Wiki Conventions
Use these conventions whenever creating or changing generated wiki pages.
## Storage boundary
The generated wiki must remain a portable tree of clean Markdown files. Do not
require a database, vector store, graph store, sidecar manifest, generated JSON,
or another persistent index to understand or retrieve it. Markdown filenames,
headings, aliases, summaries, tags, and links are the retrieval structure.
The supplied document collection is the source layer. Do not create `raw/`,
`sources/`, or per-document mirror pages in the generated wiki.
## Structure
New wikis use this minimal structure:
```text
SCHEMA.md
index.md
entities/<slug>.md
topics/<slug>.md
views/<slug>.md
```
Create a folder when its first page is needed. Do not force empty folders or
create a page for every passing mention. The first source establishes initial
knowledge, not a permanent domain boundary. Later sources may introduce
unrelated topics.
- `entities/`: people, organizations, products, projects, places, models, or
other durable named things. An entity page is the canonical home for the
subject's stable identity, not a snapshot of every observed state.
- `topics/`: concepts, domains, methods, mechanisms, definitions, recurring
ideas, and topic hubs. A topic hub may summarize and link a broader area.
- `views/`: optional cross-topic retrieval views such as comparisons,
timelines, commitments, open questions, or current-state summaries. Views
are derived reading paths: they link to canonical topic and entity pages
instead of becoming a second source of truth.
Do not create subject-specific directory trees. Represent multiple membership
with tags and Markdown links so one page can participate in several topics.
Comparisons and summaries normally belong in a topic page or an optional view,
not in dedicated top-level directories.
For an existing wiki, preserve useful legacy directories and pages. Do not
duplicate a legacy `concepts/`, `comparisons/`, or `summaries/` page under the
new structure merely to normalize it. Migrate or rename existing pages only
during an explicit reorganization when the benefit outweighs link churn.
## Semantic model
- Model the real subject described by the evidence, not a source document's
outline, section names, or storage schema. One source may update several
canonical pages, and several sources may update one page.
- Keep stable identity or definition separate from observations. Dates,
releases, measurements, changing states, forecasts, and source-specific
claims remain time-qualified in the page body rather than becoming timeless
identity fields.
- Use human-readable relationship labels such as `Part of`, `Developed by`,
`Depends on`, `Supersedes`, `Applies to`, or `Compared with` when supported
by the evidence. Use a generic related-pages list only when no more precise
relationship is known.
- Prefer composition through links, aliases, and tags over deep taxonomies or
duplicated pages. A subject may participate in several topics while keeping
one canonical page.
- Let the model evolve incrementally. Add a reusable relation label, tag, or
page convention to `SCHEMA.md` after it proves useful across recurring cases;
do not design a complete ontology from the first source.
## Foundation pages
On the first ingest, create `SCHEMA.md` and `index.md` before adding knowledge
pages. For an existing wiki, preserve useful structure and update these pages
instead of replacing them wholesale.
- `SCHEMA.md`: state that the wiki may cover multiple unrelated topics; record
this folder taxonomy, filename and link conventions, the evolving controlled
tag and alias vocabulary, page creation threshold, and any topic-specific
rules. Keep it short.
- `index.md`: a concise content router. Link major topic hubs, important
entities, and useful views with one-line summaries. Do not turn it into an
exhaustive flat inventory or maintain a page count. Every active page should
be reachable from the index through a short, meaningful link path, normally
one or two hops.
Do not create `log.md` by default. Page `updated` fields and the surrounding
revision system carry routine change history. Preserve an existing `log.md`,
but update it only when the user has chosen to maintain a human-readable audit
log.
## Page language
- Determine a source's dominant natural language from its prose, ignoring code
blocks, quotations, citations, and embedded identifiers.
- Write a new page in the dominant language of the evidence that primarily
supports it.
- When updating an existing page, preserve its established language and
integrate the new evidence coherently in that language. Do not alternate
languages paragraph by paragraph.
- When evidence uses several languages and none is clearly dominant, use the
existing page language; for a new page, use the language of the primary or
earliest authoritative source.
- Keep proper names and important technical terms in their established form.
Add useful translations, abbreviations, former names, and alternate spellings
to `aliases`.
- Do not create duplicate pages per language unless the user explicitly asks
for a multilingual edition.
- Keep `SCHEMA.md`, `index.md`, and existing views in their established
language. A first ingest uses the dominant language of the first source.
## Page conventions
- Use stable lowercase hyphenated filenames and paths, such as
`topics/attention-mechanism.md`. A filename need not be translated when the
page language later changes.
- Every generated knowledge page begins with YAML frontmatter:
```yaml
---
title: Human-readable title
aliases: [abbreviation, alternate spelling, translated name]
updated: YYYY-MM-DD
tags: [controlled, tags]
sourceIds: [stable-source-id]
---
```
- Keep frontmatter minimal. The containing directory expresses the broad page
role. Do not add page-wide confidence or contested fields when individual
claims have different evidence quality.
- `aliases` is part of retrieval. Include useful abbreviations, alternate
spellings, former names, and common translations, but not speculative or
irrelevant keyword stuffing.
- `sourceIds` records page-level provenance only. Never copy whole source
documents into the generated wiki. Cite the relevant source id close to a
specific claim when its origin, date, confidence, or conflict matters.
- Use a small number of controlled topic tags from `SCHEMA.md`. Add a durable
new tag or alias there when a later source introduces a genuinely new topic;
do not force it into the first source's taxonomy.
- Place a concise, self-contained summary paragraph immediately after the H1.
In one to three sentences, state what the subject is and why it matters. Use
the canonical name and important terminology so lexical search can find it.
- After the summary, use descriptive headings appropriate to the page's
language. Prefer focused sections for key knowledge, relationships, and
evidence or conflicts when those sections are relevant.
- On entity pages, keep durable identity or definition near the beginning.
Put volatile facts under dated status, history, observation, or evidence
sections as appropriate to the page's language and subject.
- In relationship sections, prefer short labeled links or sentences that state
how subjects relate. Do not encode relationships as opaque IDs or require a
separate graph representation.
- Link related pages using standard relative Markdown links, for example
`[Attention](../topics/attention.md)`. Index links are relative to the root,
for example `[Attention](topics/attention.md)`.
- New or meaningfully updated pages should link to related pages when such pages
exist. Do not invent links merely to satisfy a quota.
- Keep pages focused and scannable. Split a page that grows beyond roughly 200
lines into focused pages and update the links and index.
## Editorial policy
- Create a page when a topic is central to one source or recurs across sources;
otherwise merge the useful detail into an existing page.
- Before creating a page, search titles, paths, aliases, tags, and related pages
for an existing canonical home. Update that page instead of creating a
synonym or translated duplicate.
- A later source may introduce a completely unrelated subject. Add a new topic
hub or index section when it has durable value; do not invent a relationship
to existing topics.
- Keep one canonical page for a subject that spans multiple topics. Use tags
and links to represent every meaningful membership.
- Update an existing page when new evidence confirms, refines, contradicts, or
supersedes it. Update the `updated` date and provenance.
- Never silently erase a material conflict. Describe the conflicting claims
together with dates and source document ids in the page body. Keep confidence
and conflict notes close to the claims they qualify.
- Preserve useful user-authored pages and structure. Do not regenerate, rename,
or translate unrelated pages because the selected skill changed.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |