kg-skill

This skill should be used when the user asks to expose the project graph as MCP tools, register the graphify MCP server (python -m graphify.serve) in Claude config, says 'graphify mcp 서버', 'MCP로 그래프 쿼리', or invokes /kg-mcp [register|status|unregister]. Configures Claude's .mcp.json so subsequent /kg-query and /kg-orient can call graphify MCP tools instead of loading graph.json directly. Edits user config — requires explicit approval.

This skill should be used at the start of a session when the project contains wiki/ or graphify-out/, when the user asks '현재 kg 상태 보여줘', '세션 시작 요약', 'what is the current wiki state?', or invokes /kg-orient. Reads hot.md, overview.md, graph reports, freshness — read-only, never modifies files.

Router for the Knowledge Graph + LLM Wiki system. Use when the user invokes /kg explicitly or asks 'which kg command should I use?'. Sub-commands — /kg-orient, /kg-update, /kg-query, /kg-lint, /kg-ingest (data ops) and /kg-reflect, /kg-challenge, /kg-connect, /kg-suggest (dialogue ops). Project-state auto-trigger (wiki/ or graphify-out/ presence) belongs to /kg-orient, not this router.

This skill should be used when the user asks to visualize the kg wiki, create an Obsidian Canvas, knowledge map, graph view, mindmap, community map, or invokes /kg-canvas. Exports read-only .canvas JSON views — never modifies content pages.

This skill should be used when the user asks to initialize, bootstrap, reset, or prepare a kg-skill wiki, says 'kg-skill 처음 설정', 'wiki 초기화', '스키마 pin 만들기', or invokes /kg-init. Creates wiki/, copies schema, seeds index/log/hot, validates setup. Idempotent — never overwrites existing wiki content.

This skill should be used when the user asks to audit, validate, health-check, repair, or lint the kg wiki/graph, says '위키 건강 점검', '깨진 링크 찾아줘', 'orphan page 확인', or invokes /kg-lint. Reports graph integrity, freshness, missing links, stale claims, schema proposal debt. With --fix applies only safe bookkeeping changes.

This skill should be used when the user asks to merge graphs across multiple projects, build a cross-repo knowledge graph, says '여러 프로젝트 통합 그래프', 'cross-project KG', 'merge graphs', or invokes /kg-merge. Reads each project's graphify-out/graph.json and produces a unified graphify-out/merged-graph.json via graphify v0.8.x merge-graphs. Read-only on source projects.

Query accumulated knowledge — BFS/DFS graph traversal + wiki lookup. Supports --depth quick|standard|deep for token-efficient queries. Use when asking questions ABOUT existing knowledge (relationships, evidence, decisions). For PROPOSING new connections between communities, use /kg-connect instead.

This skill should be used when source files changed and the user asks to rebuild, refresh, or update the structural graph, says '그래프 갱신', 'Graphify 업데이트', 'changed files 반영', or invokes /kg-update. Delegates to graphify update; writes graphify-out/ and wiki/graph-report*.md, and maintains the wiki/index.md Graph snapshot section + wiki/log.md audit entry. Never modifies schema or content wiki pages.

This skill should be used when the user asks to find missing links between graph communities, bridge isolated concepts, connect two topics, or invokes /kg-connect. Proposes inferred semantic bridges with rationale and risk — never adds edges without user confirmation.

Ingest source files into the wiki — read, discuss key takeaways, write source/entity/concept pages with wikilinks. Requires wiki/ to be initialized (run /kg-init first).

This skill should be used after multiple ingests, at session end, when the user asks for hidden tensions, emerging patterns, blind spots, schema drift, or invokes /kg-reflect. Surfaces only non-obvious insights supported by multiple wiki/graph signals — explicitly says 'no novel insights' if none.

Autonomous multi-round web research loop (자율 웹 조사/자료조사/문헌 검토). Decomposes a topic into angles, searches, fetches, extracts findings, and queues results for user approval before filing into the wiki. Governed by research-policy.yaml. Use when /kg-query --depth deep says "insufficient wiki coverage" or 새로운 주제를 체계적으로 조사할 때.

This skill should be used when the user asks to stress-test, challenge, falsify, or argue against a wiki claim, decision, heuristic, or assumption, says '반박해줘', '이 주장 검증해줘', 'devil's advocate', or invokes /kg-challenge. Uses only evidence already in the knowledge base — never fabricates doubt.

This skill should be used when the user mentions tacit knowledge, repeated mistakes, gotchas, lessons learned, 비법, 주의사항, 실패 경험, or invokes /kg-elicit. Captures implicit knowledge as Experience first, optionally promotes to Heuristic only after explicit confirmation.

This skill should be used after a recent discrete failure, build error, unexpected result, debugging session, 삽질, or '이걸 기록해뒀어야 했는데' moment, or when the user invokes /kg-postmortem. Captures one trial-and-error event as Experience and optionally Heuristic only if user confirms generalization.

Evolve the kg ontology (온톨로지 진화/스키마 변경/분류 체계 수정). Sub-commands propose/approve/migrate/diff/list/pull-global. Changes are proposal-first; nothing is auto-applied. Per-wiki pin is default target; use --global to modify the shipped default.

This skill should be used when the user asks what to read, ingest, research, or improve next in the kg wiki, says '다음에 뭘 봐야 해?', '지식 공백 찾아줘', or invokes /kg-suggest. Recommends 3-5 concrete next sources or maintenance actions based on unresolved links, stale topics, low-cohesion communities.