gnapsis: Skill for Claude Code

Build Knowledge Graph

Reverse engineer the architecture of the codebase and build a comprehensive knowledge graph using the gnapsis MCP tools, following a strict derivation order methodology.

If an argument is provided, focus the analysis on that scope or path. Otherwise, analyze the entire codebase.

---

CRITICAL RULES

1. Always use gnapsis MCP tools to register all nodes and relationships in the knowledge graph. The gnapsis graph is the primary output.
2. Proceed through phases, summarize at each phase boundary. Complete each phase fully, then summarize what was done before moving to the next.
3. Use best judgment for ambiguity: When you encounter ambiguity about business domains, feature boundaries, or architectural decisions, use your best technical judgment and document your reasoning. Add a note in the entity description when a decision was ambiguous.
4. ALWAYS run analyze_document before creating entities for a source file. This gives you the exact LSP symbol names. Never guess symbol names.
5. Use ref_type: "code" with lsp_symbol for source files. Only use ref_type: "text" for markdown, docs, and config files.
6. Be exhaustive, not superficial. Scan every directory, every configuration file, every module entry point. Leave no stone unturned.

---

GNAPSIS TOOL REFERENCE

Initialization & Discovery

• init_project — Initialize the database schema (run once at start)
• project_overview — Get current ontology: taxonomy (categories by scope), entity hierarchy, statistics

Entity Lifecycle

• create_entity(name, description, category_ids, parent_ids, commands) — Create an entity with at least one reference
• update_entity(entity_id, ...) — Update entity, add/remove references, create relationships
• delete_entity(entity_id) — Delete entity (must have no children)

Taxonomy

• create_category(name, scope) — Create a new category at a scope (if the defaults don't fit)

Querying

• get_entity(entity_id) — Full entity details with references and relationships
• find_entities(scope, category, parent_id) — Filter entities by scope/category/parent
• search(query) — Semantic search across entities and references
• query(entity_id, semantic_query) — Extract relevant subgraph within token budget
• get_document_entities(document_path) — Get all entities referenced in a file