Codegraph

CodeGraph for MOOSE

CodeGraph is a pre-built knowledge graph of every symbol, edge, and file in MOOSE. It answers "does this already exist?" and "what does this touch?" in one sub-second call - exactly what you need to honor MOOSE's Simplicity-First rule ("if existing functionality can accomplish a subtask, use it, don't reimplement it; if it's nearly capable, prefer a short extension over a long rewrite") and something grep cannot do reliably at this scale.

Prefer the MCP tools (codegraph_explore, codegraph_search, codegraph_node, codegraph_callers). If they aren't loaded, the shell equivalents print the same output: codegraph explore "...", codegraph node <name>, codegraph query <name>, codegraph callers <name>, codegraph impact <name>.

Availability (optional - do not block on it)

Confirm .codegraph/ exists at the repo root, or run codegraph status (it prints the file/ node/edge counts). If it is missing, CodeGraph is not set up for this repo:

• Fall back to grep/Read for the same questions - do not stop, and
• Suggest setting it up once so future sessions are faster and more reliable:

codegraph install   # one-time: register the CodeGraph MCP server with this agent
codegraph init      # build the index for this repo (run from the repo root; takes a few minutes)

The index lags writes by ~1s via a file watcher, so it reflects the working tree.

(The pr-review skill requires CodeGraph and stops without it. General coding does not - this is the soft entry point.)

Primary use: don't reinvent existing functionality

Before you add a new function, class, or helper, check whether MOOSE already provides it.

1. Name the symbol you are about to write - the function/class/method and what it does.

2. Ask whether MOOSE already has it, from two angles, because a reinvented helper is often named differently than the original:

• By name/near-name: codegraph_search the name and obvious variants (a new computeNormalVector -> search normal, normalize, unitVector).
• By concept: codegraph_explore a short description of what the code does ("interpolate value at quadrature point", "parse comma separated list to vector", "L2 norm of a vector"). This returns the verbatim source of the closest existing symbols, so you can compare behavior directly.

3. Act on the finding. If an existing symbol does the same thing, reuse it - or extend/fix it if it's nearly capable (a short extension is preferred over a long reimplementation). Only write new code if the existing one genuinely differs (different semantics, performance contract, or it is deprecated).

Common MOOSE reinvention hot-spots: math/tensor helpers (often already in libMesh, MetaPhysicL, or MathUtils), string/parsing utilities (MooseUtils), mesh/geometry queries, and "generic" materials/functors that mirror an existing templated one.

Note: the default codegraph.json excludes the libmesh and contrib trees (so MetaPhysicL too) to keep the index small and fast, so CodeGraph will not surface a helper reinvented from those. For the libMesh/MetaPhysicL hot-spots, fall back to grep or prior knowledge - or index those trees explicitly when the check matters for a given PR.

Other uses

• Blast radius before you change a shared symbol. codegraph_callers / codegraph impact lists every call site, so you can update them all and gauge how risky the change is.
• Follow the established pattern. Before writing a new object, codegraph_explore a sibling that already does something similar (a new Kernel -> an existing comparable Kernel) and match its structure unless there's a reason not to.
• Verify a claim. When a comment or PR says "this mirrors X" or "replaces Y", pull X/Y with codegraph_node (includeCode=true) and confirm it instead of taking it on faith.

Practical notes

• codegraph_explore is the workhorse - one call returns relevant symbols' source grouped by file plus the call paths between them, so it usually answers the question without follow-up Reads. Treat its output as already-Read.
• For an ambiguous name (many overloads/same-named classes across modules), codegraph_node returns every matching definition's body; narrow with the file argument.
• Keep queries specific. Bag-of-symbols ("MathUtils pow raisePower") and natural-language questions both work.