Wiki Orient (Dexio)
Read an LLM wiki before working in it or answering from it: the schema page, the catalog of page descriptions, recent changes, then a targeted search and the canonical page for the…
Dexio
@dexio
Install
$ openclaw skills install @dexio/dexio-wiki-orientOrient in a wiki
Most wiki damage comes from writing before reading: a duplicate page because the agent did not search, a convention broken because it did not read the schema, a fact overwritten because it did not see that another agent changed the page an hour ago. Orienting costs a few tool calls and prevents all three.
Do this once per task, not once per edit. Do not read the whole wiki.
Steps
- Read the schema.
SCHEMA.md(orAGENTS.md,CLAUDE.md, or the page the wiki's README points to). It says where pages go, what frontmatter they carry, which tags exist, and when to create a page instead of editing one. The wiki's own schema overrides the defaults in these skills. - Scan the catalog. Folder wiki: list the files, or run
python3 <wiki-lint skill>/scripts/wiki_lint.py <wiki> --catalogto get every page with its one-line description. Hosted wiki over MCP:list_pages. You are looking for which folders exist and which pages sound like your topic, not reading them. - Check recent changes. Folder in git:
git log --since="7 days ago" --stat -- <wiki>. Hosted: the history tool without a path (on Dexio,page_history). Plain folder with no history: the most recently modified files (find <wiki> -name '*.md' -mtime -7). This shows what other agents are working on, so you do not redo or undo it. - Search the topic. Two or three phrasings: the entity's name, the concept, and the words
a person would use. Folder:
grep -rilor ripgrep across the wiki. Hosted:search_pages. Search before concluding that something is not in the wiki. - Read the canonical page, then its neighbors. The page that owns the topic, and the one or two pages it links to that bear on your task. Stop there unless the task needs more.
What to carry forward
- The path of the page that owns your topic. If none exists, note that too;
wiki-recorddecides whether one should. - The page's
updateddate and, on a hosted wiki, its version, so a later write can detect that someone else changed it in between. - Any claim you will rely on that is about a live system (a price, a config value, an account's state, a person's current role). The wiki records what was true when written. Check the original source before acting on it.
Pitfalls
- Treating the wiki as the system of record. It is curated memory. For anything mutable, it tells you where to look and what was true then.
- Reading a hand-kept
index.mdinstead of listing pages. Index files drift; the file listing or the host's page list does not. - Loading every page "for context". It wastes the budget you need for the task and makes you more likely to repeat stale claims.
- Asking a person for context before searching. If the wiki holds it and you ask anyway, you have made the wiki pointless for them.
Top skills in this category
self-improving agent
@pskoettCaptures and maintains learnings, errors, and corrections. Use when a command fails, the user corrects an assumption, a capability is missing, knowledge is outdated, or a better approach is found. Review relevant saved guidance before major tasks, reuse, and promotion; revalidate it according to wha
Self-Improving + Proactive Agent
@ivangdavilaSelf-reflection + Self-criticism + Self-learning + Self-organizing memory. Agent evaluates its own work, catches mistakes, and improves permanently. Use when...
ontology
@oswalpalashTyped knowledge graph for structured agent memory and composable skills. Use when creating/querying entities (Person, Project, Task, Event, Document), linkin...
Humanizer
@biostartechnologyRemove signs of AI-generated writing from text. Use when editing or reviewing text to make it sound more natural and human-written. Based on Wikipedia's comprehensive "Signs of AI writing" guide. Detects and fixes patterns including: inflated symbolism, promotional language, superficial -ing analyses, vague attributions, em dash overuse, rule of three, AI vocabulary words, negative parallelisms, and excessive conjunctive phrases.
Obsidian
@steipeteWork with Obsidian vaults (plain Markdown notes) and automate via obsidian-cli.