graphwiki

name	graphwiki
version	2.0.0
trigger	/graphwiki
description	LLM knowledge graph with persistent wiki compilation
platforms	["claude","codex","auggie","gemini","cursor","openclaw","copilot","windsurf","cody","codewhisperer","opencode","aider","droid","trae","trae-cn","antigravity","hermes"]

GraphWiki Skill

You have access to the GraphWiki knowledge graph for persistent, token-efficient context loading.

What To Do When Invoked

When /graphwiki is triggered, follow this sequence:

1. Check if a graph exists — look for graphwiki-out/graph.json (or .graphwiki/graph.json).
2. If the graph exists — load it and answer the user's question based on graph nodes and edges. Do not rebuild unless the user explicitly asks.
3. If no graph exists — inform the user and offer to run graphwiki build . --update to create one.
4. Run queries against the loaded graph — use graphwiki query for semantic questions, graphwiki path for structural relationships (0 LLM tokens).
5. Return structured context — summarize relevant nodes, edges, and wiki content relevant to the user's question.

Context Loading Protocol

graph exists?
  YES → load graphwiki-out/graph.json → query or path → return structured context
  NO  → warn user → offer: graphwiki build . --update

When loading graph context manually (hook unavailable):

1. Load graph overview from graphwiki-out/graph.json (nodes + edges summary)
2. Run graphwiki path <nodeA> <nodeB> to find structural relationships
3. Check wiki index at graphwiki-out/wiki/
4. Load up to 3 wiki pages maximum per query (token budget)
5. Fall back to raw source files only if graph data is insufficient

Key Commands

Command	Purpose	Notes
graphwiki build . --update	Incremental rebuild (changed files only)	Preferred over full rebuild
graphwiki build .	Full graph rebuild from scratch	Use only when graph is stale or missing
graphwiki build . --watch	Watch mode — auto-rebuild on file changes	For active development sessions
graphwiki build . --directed	Build directed graph (preserves edge direction)	Use for dependency/flow analysis
graphwiki build . --mode deep	Deep extraction mode via LLM	More thorough, higher token cost
graphwiki query "question"	Semantic query against knowledge base	Returns relevant wiki content
graphwiki ask "question"	LLM-augmented question answering	Uses graph + LLM reasoning
graphwiki add <url>	Add URL source to the graph	Fetches and indexes remote content
graphwiki explain <node>	Explain a node's role and connections	Useful for onboarding
graphwiki path <nodeA> <nodeB>	Find structural path between two nodes	Zero LLM tokens
graphwiki save-result <file>	Persist a query result to disk	For caching expensive queries
graphwiki hook install	Install PreToolUse/PostToolUse hooks	Enables automatic context injection
graphwiki hook uninstall	Remove hooks
graphwiki hook status	Check hook registration status
graphwiki skill uninstall --all	Remove all platform skill installations
graphwiki lint	Validate graph integrity	Run before committing graph changes
graphwiki status	Show graph stats and token budget usage

Pending Extraction Prompts

GraphWiki generates extraction prompts for non-code files (PDFs, images, binary assets). Check .graphwiki/pending/ for any .prompt.md files before answering.

Protocol:

1. Check .graphwiki/pending/ for *.prompt.md files
2. Read each prompt — it describes what knowledge to extract and the expected output schema
3. Extract the requested knowledge
4. Write the result JSON to the path specified in the prompt
5. Run graphwiki build . --update to merge the extracted knowledge into the graph

If .graphwiki/pending/ is empty or does not exist, skip this step.

Hard Constraints

• NEVER modify source files — treat all files in raw/ and source directories as immutable
• NEVER manually edit graphwiki-out/ — all output is auto-generated; changes will be overwritten
• Always check the existing graph first before suggesting a rebuild — incremental --update is almost always sufficient
• Use --update for incremental builds — full rebuilds are expensive and rarely needed
• Cap wiki page loads at 3 per query — respect the 150K session token budget (hook warns at 80%)
• Zero-token path queries — always prefer graphwiki path over LLM reasoning for structural questions

Agent Role Matrix

Agent Role	Recommended GraphWiki Usage
codebase-search	graphwiki path <term1> <term2> — 0 tokens
requirements	graphwiki query to load relevant wiki pages
planning	graphwiki status and graphwiki lint
implementation	graphwiki build . --update after file changes
verification	graphwiki lint and graphwiki status

References

• references/commands.md — Full command table with all flags and examples
• references/platform-install.md — Platform-specific installation and configuration

Generator

skill-generator.ts parses this file and generates platform-specific skills: SKILL-claude.md, SKILL-codex.md, SKILL-copilot.md, SKILL-auggie.md, SKILL-gemini.md, SKILL-cursor.md, SKILL-openclaw.md, SKILL-windsurf.md, SKILL-cody.md, SKILL-codewhisperer.md, and others.