/graph

Run a graph-layer operation against the wiki using the llm-wiki skill's graph workflow.

Arguments: $ARGUMENTS

1. Confirm wiki/graph/ontology.yaml exists. If not, the wiki is pre-graph — offer to seed wiki/graph/ (copy the templates from skills/llm-wiki/assets/) before doing anything else. Do not fabricate ontology entries.
2. Dispatch on the first argument:
   • extract → uv run --script "${CLAUDE_PLUGIN_ROOT}/skills/llm-wiki/scripts/wiki_graph_extract.py" "${PWD}/wiki"
   • lint → uv run --script "${CLAUDE_PLUGIN_ROOT}/skills/llm-wiki/scripts/wiki_graph_lint.py" "${PWD}/wiki"
   • neighbors --node <id> → python "${CLAUDE_PLUGIN_ROOT}/skills/llm-wiki/scripts/wiki_graph_query.py" "${PWD}/wiki" neighbors --node <id>
   • edges --subject <id> [--predicate <p>] → python "${CLAUDE_PLUGIN_ROOT}/skills/llm-wiki/scripts/wiki_graph_query.py" "${PWD}/wiki" edges --subject <id> [--predicate <p>]
   • path --from <id> --to <id> [--max-depth N] → python "${CLAUDE_PLUGIN_ROOT}/skills/llm-wiki/scripts/wiki_graph_query.py" "${PWD}/wiki" path --from <id> --to <id>
   • facts --about <id> → python "${CLAUDE_PLUGIN_ROOT}/skills/llm-wiki/scripts/wiki_graph_query.py" "${PWD}/wiki" facts --about <id>
3. For query subcommands, after running the script, follow the most relevant edges back to their wiki pages and summarize. Cite with [[wikilinks]] to wiki pages, not graph rows. The graph accelerates navigation; the wiki page and its raw source remain the evidence.
4. For lint, present findings as proposed edits — do not silently rewrite typed graph.relationships. After approved fixes, suggest re-running extract so the compiled artifacts stay in sync with the markdown.
5. If graph.sqlite is older than the most recent ingest line in log.md, suggest running extract first.

Full reference: skills/llm-wiki/references/graph-workflow.md.