Dependency graph

The graph is parsed, never written by hand or guessed. scripts/graph.mjs runs dependency-cruiser over the code, with the project's own TypeScript, and turns the ask into options. The same options give the same graph on the same commit.

Run it

Translate the ask into options, then run it once:

node <skill-dir>/scripts/graph.mjs --root <the package's folder> [--from <path>] [--to <path>] [--focus <path> --depth <n>] [--steps <n>] [--level <n>] [--only <regex>] [--exclude <regex>] --title "<title>" --out <path>

• What a module depends on: --from that module or folder.
• Who depends on it: --to it.
• Its neighbourhood: --focus it, --depth the hops.
• Not too far: --steps <n> follows imports only n steps from --from.
• Folders, not files: --level 2 folds src/<area>/… into one node per area; the coarser view first, then open the part that matters.
• Noise out: --exclude "tests|generated"; node_modules and Node's own modules are always out.

The root is the folder holding the package's tsconfig.json (its extends resolves from there); in a monorepo, one package per graph. The first run installs dependency-cruiser and the project's TypeScript into a cache, needing the network once.

The document

Through the aiview skill: kind report from YYYY-MM-DD-<subject>.dependency-graph.report.md, tags dependency-graph and the module, group the piece of work it serves. The script writes it at the path aiview path gives, with the commit, the counts and the command that makes it again. Open it and tell the person the URL. A graph over about a hundred nodes is folded with --level or narrowed before it is shown.

What the graph shows is said under it, in a few lines: the cycles, a dependency against the project's stated boundaries (its conventions file), a module far more depended on than the rest. Nothing beyond what the edges show.