Skip to content
T1KTheOneKit

t1k:graphify

FieldValue
Modulet1k-maintainer
Version3.2.1
Efforthigh
Tools—

Keywords: ast, code-analysis, codebase-understanding, graphify, knowledge-graph, tree-sitter

How to invoke

Section titled “How to invoke”
/t1k:graphify
[path] [--mcp|--report|--watch]

Graphify — Knowledge Graph Builder

Section titled “Graphify — Knowledge Graph Builder”

Turn any folder of code, docs, papers, or images into a queryable knowledge graph. Uses tree-sitter AST for code (20 languages), Whisper for audio/video, and LLM subagents for documents.

Pre-flight Step 0 — Fuzzy plan/path arg resolution (MANDATORY)

Section titled “Pre-flight Step 0 — Fuzzy plan/path arg resolution (MANDATORY)”

If the user provides a fuzzy plan/path/phase arg (e.g. chaosforge-demo, plans/chaosforge-demo, phase-3), an empty arg, or natural-language ref like “active plan” / “current plan” / “this plan”, run the Fuzzy Plan / Path Resolution Protocol at skills/t1k-cook/references/fuzzy-plan-resolution.md BEFORE bail. Skill MUST NOT emit “no path matching” / “exact path required” until that protocol has been applied and Step 6 reached.

When to Use

Section titled “When to Use”
  • Understanding unfamiliar codebase architecture before planning
  • Discovering cross-file relationships and dependency chains
  • Finding “god nodes” (most-connected concepts) in large projects
  • Navigating by structure instead of grepping every file
  • Preparing context-efficient codebase representation (71.5x fewer tokens vs raw files)

Typically precedes: /t1k:plan (understand architecture before planning) Related: /t1k:scout (quick file search), /t1k:repomix (full context dump)

Quick Start

Section titled “Quick Start”

This is an optional skill wrapping the third-party graphifyy package (Python 3.10+; PyPI name is double-y — other graphify* packages are unaffiliated). Install + full CLI/MCP/architecture detail: references/reference.md.

Terminal window
pip install graphifyy && graphify install   # install + download tree-sitter grammars


graphify .                    # build knowledge graph from current directory
graphify /path/to/project     # or a specific path
graphify . --watch            # watch mode, auto-rebuild on file changes

Output lands in graphify-out/ — graph.html (interactive viz), GRAPH_REPORT.md (god nodes, suggested questions), graph.json (persistent graph), cache/ (incremental rebuilds).

Tool Size Caps (E6)

Section titled “Tool Size Caps (E6)”

When invoking the graphify MCP server’s query_graph and get_neighbors tools, this skill MUST pass an explicit maxResultSizeChars cap to prevent context blow-up on large graphs:

  • query_graph: cap at maxResultSizeChars: 200_000 (~50k tokens). Larger results break Claude’s working memory; instead, paginate via cursor.
  • get_neighbors: cap at maxResultSizeChars: 50_000 (~12k tokens). Neighbor expansion can fan out exponentially in dense graphs.
  • Always include limit AND maxResultSizeChars — limit bounds nodes, but a single node with megabyte-sized properties still blows the budget.

If a query exceeds the cap, the MCP server returns truncated: true; the skill MUST surface this to the user with a hint to refine the query, NOT silently deliver a partial result as if it were complete.

Gotchas

Section titled “Gotchas”
  • Beta status: depends on a third-party package (graphifyy); API surface may change.
  • Privacy: code is processed locally via tree-sitter AST and audio/video locally via Whisper — nothing leaves the machine for those. Docs/images ARE sent to your configured model provider (Claude/OpenAI) for semantic extraction.
  • First build on large codebases can be slow (AST parsing + LLM calls); semantic-extraction quality depends on the underlying model.
  • Full architecture (three-pass extraction, confidence tagging, MCP tool list, language coverage): references/reference.md.