Menu
First-time Loom setup for a new repository. Runs loom analyze, installs MCP config, and shows what was indexed.
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Baseado na classificação ocupacional SOC
Read undocumented functions and store one-sentence summaries into Loom. Prioritizes hot functions and missing-summary suggestions. Makes every future session faster.
Navigate and understand a codebase using Loom context packets. Find functions, trace call chains, understand architecture — without reading files.
Assess change risk before modifying a function. Maps blast radius, surfaces hidden dependencies, community cohesion, and unexpected cross-module connections.
Use Loom MCP tools for code intelligence — symbol search, context packets, blast radius, session primer, delta context, topology insights. Invoke when working on any codebase indexed with `loom analyze .`.
| name | onboard |
| description | First-time Loom setup for a new repository. Runs loom analyze, installs MCP config, and shows what was indexed. |
| argument-hint | [repo-path] |
| allowed-tools | ["mcp__loom__graph_stats","mcp__loom__god_nodes","mcp__loom__suggest_questions","mcp__loom__get_savings","Bash","Read","Glob"] |
Set up Loom for a repository and orient to the codebase.
Call graph_stats() — if total node count > 0, Loom is already indexed. Jump to Step 4.
loom analyze .
This runs tree-sitter over all Python/TypeScript/JavaScript/Java files and builds ~/.loom/projects/{repo-name}.db (or ~/.loom/loom.db if not in a git repo). Typically 5–30 seconds for a 500-file repo.
If the user provides a path argument, use that instead of ..
loom install
This writes the MCP server config to the appropriate tool config file (Claude Code, Cursor, Windsurf, or Codex), and installs a post-commit git hook for incremental sync.
Call graph_stats() — shows node and edge counts by kind.
Call god_nodes(limit=10) — shows the most-called functions. These are the best starting points for understanding the codebase.
Call suggest_questions(limit=7) — surfaces dead code, undocumented hot functions, bridge nodes, and low-cohesion clusters worth looking at.
Show: