メニュー
Get oriented in a codebase before reading files — repo shape, entry points, where the graph is blind, and the full structural map. Use at the start of any task in an unfamiliar or large repo.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Start from a failure, not a filename — turn a stacktrace, failing test, or diff into the exact code that matters, then read it. Use when you have an error and need to know where to look.
Locate where a feature or symbol lives — match nodes by name, or pull the related cluster from a seed via spreading activation. Use to scope before reading.
Blast radius before a change — everything a node affects (forward) or that depends on it (backward), ranked and located, with likely-dead code flagged. Also the one-hop view of direct connections.
Set up or rebuild the structural graph for a repo — on first use, after a big refactor, or to wire repo-graph into a project permanently.
Follow the code across boundaries — a feature end-to-end, or the shortest path between two nodes (e.g. a frontend action to its backend handler). Links across the stack in one call instead of grepping both sides.
| name | repo-graph-orient |
| description | Get oriented in a codebase before reading files — repo shape, entry points, where the graph is blind, and the full structural map. Use at the start of any task in an unfamiliar or large repo. |
Build a mental model of the repo from the graph instead of opening files one by one.
orient first — node/edge counts, detected kinds, entry points, and a blind spots note: which languages/edge-types the graph under-links, so you know exactly where to fall back to grep. This is the orient call; do it before grepping.orient with full=true — the full graph in dense notation. Read it instead of reading files; it's the primary context source. Large graphs are truncated, so scope with orient <node> or find if needed.orient with seed=<node> — the scoped map around it.Report the tiers (entry → service → handler → data) and the main entry points. You now know where to look without having opened a file.