| name | knowledge-graph |
| description | Read and refresh the project's dependency graph to trace classes, services, and modules. |
Knowledge Graph
A pre-built dependency graph lives at .claude/graph/dependency-graph.json. The
builder is node bin/build-knowledge-graph.js (incremental by default, --full to force
rebuild).
This skill has two responsibilities:
- Refresh the graph at session start if it is stale (
base_commit ≠ git rev-parse HEAD).
- Read the graph to answer dependency, namespace, and structure questions instantly.
Read it before running grep/glob searches for class relationships, namespace
exploration, or dependency tracing. It eliminates redundant file scans and speeds up the
first useful response in any session.
Graph shape
{
"generated_at": "<ISO timestamp>",
"base_commit": "<git SHA>",
"node_count": 913,
"nodes": {
"src/Engine/Cache/Subscriber.php": {
"language": "php",
"namespace": "Plugin\\Engine\\Cache",
"symbols": [
{ "kind": "class", "name": "Subscriber", "extends": [], "implements": ["Subscriber_Interface"] }
],
"imports": [
"Plugin\\Event_Management\\Subscriber_Interface",
"Plugin\\Engine\\Cache\\Purge"
]
}
},
"symbol_index": {
"Plugin\\Engine\\Cache\\Subscriber": "src/Engine/Cache/Subscriber.php"
}
}
nodes — keyed by relative file path. Each node has the language (php or js), declared symbols (PHP only), and all import/use statements.
symbol_index — maps every fully-qualified PHP class / interface / trait / enum to its file path. Use this for instant "where is this class?" lookups.
Query patterns
Find a class file (zero grep)
symbol_index["Plugin\\Engine\\Cache\\Purge"]
→ "src/Engine/Cache/Purge.php"
Find the ServiceProvider that wires a class
The ServiceProvider that registers a class imports it. Search for files whose imports contain the target FQN:
filter nodes where "Plugin\\Engine\\Cache\\Purge" ∈ node.imports
→ "src/Engine/Cache/ServiceProvider.php"
Then read that ServiceProvider to see how the class is registered in register().
Find all Subscribers in a module
Filter nodes where:
namespace starts with the module prefix (e.g. Plugin\Engine\Cache)
symbols[*].implements contains Subscriber_Interface
Find all ServiceProviders in the codebase
Filter nodes where symbols[*].extends contains AbstractServiceProvider.
Trace a class's full dependency chain
- Start at
symbol_index["Plugin\\...\\ClassName"] → get file path
- Read
nodes[file].imports → these are its direct dependencies
- For each dependency, repeat → you get the full constructor injection tree without reading any PHP
Verify no unexpected cross-module dependencies
Check nodes[file].imports for any FQN that shouldn't be there.
For example, a Frontend Subscriber importing an Admin class is a red flag.
Keeping the graph fresh
The graph records the git commit it was built from (base_commit). If that SHA differs from HEAD, run:
node bin/build-knowledge-graph.js
The script is incremental — it only re-parses files changed since base_commit. Use --full to force a complete rebuild.
When to refresh:
- At the start of every issue workflow session.
- After merging a branch with structural changes (new classes, namespace moves).
- Before an architecture review session.
Supported languages
| Language | What is extracted |
|---|
| PHP | namespace, class/interface/trait/enum declarations (with extends/implements), use imports (including grouped \{A, B} forms) |
| TypeScript / JavaScript | import (static + dynamic) and require() sources |
Practical workflow (issue implementation)
Before writing a single line of code for an issue:
- Check
base_commit vs HEAD — refresh if stale.
- Use
symbol_index to locate all classes involved in the fix.
- For each class, read
nodes[file].imports — know the dependency chain before touching the constructor.
- Find the ServiceProvider via the import search above — know where to add/modify the binding.
- List all Subscribers in the module — know which ones may need new hook entries.
- Only then open the actual PHP files (now you know exactly which ones to read).