Commands

Seven /wiki:* slash commands ship with the Claude Code plugin. In other agents, phrase the same requests in natural language — the underlying skill triggers either way.

Command	What it does
/wiki:init	Bootstrap a new wiki structure in the current project.
/wiki:ingest <source>	Process a new source into the wiki; refreshes the graph layer when present.
/wiki:query <question>	Answer a question from the wiki with citations.
/wiki:lint	Structural and semantic health check.
/wiki:stats	Show size, link density, and which scaling threshold the wiki is at.
/wiki:graph <action>	Compile, lint, or query the typed graph layer.
/wiki:upgrade	Upgrade an existing wiki to the current plugin version.

/wiki:init

Bootstrap a new LLM Wiki in the current project. Optional --wiki-dir / --raw-dir arguments override the default directory names.

/wiki:init

The command confirms paths, bootstraps the wiki, then installs and verifies pinned FastEmbed, sqlite-vec, and PyYAML dependencies through uv, caches the local embedding model, builds the parse cache, and embeds every current section. It walks through SCHEMA.md and proposes agent-memory integration only after runtime JSON reports "status": "ready". It does not ingest a source.

/wiki:ingest <source>

Ingest a source (paper, article, transcript, PDF, notes) into the wiki using the skill's ingest workflow.

/wiki:ingest raw/attention-is-all-you-need.pdf

The agent reads SCHEMA.md, places the raw source under raw/, chunk-reads if it is large, discusses key takeaways, surveys the wiki for touched pages, writes the source-summary page, makes surgical str_replace updates, creates new pages for new entities and concepts (each with an inbound link), updates the index, and appends one line to log.md. If the graph layer is present and the source supports typed edges, it lints and re-extracts the graph. Full detail in Workflows → Ingest.

/wiki:query <question>

Answer a question against the wiki with citations.

/wiki:query How does attention scale with sequence length?

The agent reads the index (or relevant shard), follows selected [[wikilinks]], and synthesizes an answer with citations. If index summaries are insufficient it runs local hybrid search:

uv run --script skills/llm-wiki/scripts/wiki_search.py \
  "<query terms>" --top 10 --cache --json

FastEmbed + sqlite-vec semantic ranking is fused with BM25; no API key or outbound wiki/query text is involved. For dependency-free lexical search, bypass PEP 723 resolution with python skills/llm-wiki/scripts/wiki_search.py "<query>" --no-embed; use --granularity page for whole-page ranking. See Search & retrieval. Relational questions can also consult graph.sqlite; substantive answers can be filed back into wiki/synthesis/.

/wiki:lint

Run a structural and semantic health check. Add --suggest-pages to surface concepts that should be promoted to their own page.

/wiki:lint

The structural pass runs wiki_lint.py to catch orphans, broken wikilinks, oversized pages, missing frontmatter, stale pages, and duplicate slugs. The semantic pass reads recently-updated pages and hub pages for contradictions and gaps. Every finding is presented as a proposed edit — nothing is applied silently. Full detail in Workflows → Lint.

/wiki:stats

Show wiki size, shape, and link density — and which scaling threshold the wiki is at.

/wiki:stats

Runs wiki_stats.py and interprets the output: whether you are approaching a sharding threshold (150 pages / 300-line index), the routine-search threshold (300 pages), or a weekly-lint cadence (500 pages); whether any pages exceed the 400-line soft cap or 800-line hard cap; and whether link density looks low.

/wiki:graph <action>

Compile, lint, or query the wiki's typed graph layer. The graph is optional and requires wiki/graph/ontology.yaml.

/wiki:graph extract
/wiki:graph neighbors --node product:konvy
/wiki:graph edges --subject person:stephanie-emmanouel
/wiki:graph path --from person:praney-behl --to product:konvy
/wiki:graph facts --about product:konvy

extract compiles the markdown into nodes.jsonl, edges.jsonl, graph.sqlite, and graph.graphml; lint validates typed edges; the query subcommands navigate the compiled graph. Markdown stays canonical — the graph can be deleted and rebuilt at any time. See the Graph layer guide.

/wiki:upgrade

Upgrade an existing wiki with idempotent file operations, mandatory runtime installation/index synchronization, and a walked SCHEMA.md merge.

/wiki:upgrade

Resolves the installed init_wiki.py separately from the current project's absolute root, then runs it with --upgrade. Existing content stays untouched while the complete pinned local runtime is installed and verified, parse/vector indexes synchronize, and missing SCHEMA.md sections are walked one at a time. See Upgrade to v3.