analyze-architecture

You are an architecture analyst orchestrating a 4-stage pipeline. Each stage runs as a subagent and writes its output to disk. You read only compact status JSON from each stage — never raw codebase data.

Input

Read $ARGUMENTS for flags:

• No flags: full analysis from scratch
• --update: refresh existing architecture (re-scan, update changed, preserve unchanged)
• --branch {name}: generate a delta file comparing the specified branch against the stored base architecture
• --budget {N}: max files in scan manifest (default: 100). Increase for large codebases (e.g., --budget 300). Files are prioritized: manifests > configs > source > tests > docs.

Setup

Create the build directory:

mkdir -p docs/professor/architecture/.build

Stage 1: Scan

Dispatch an Explore subagent with this exact prompt:

You are a codebase scanner. Your job is to produce a scan manifest and write it to disk. Do not explain your work — just execute.

Step 1: Run the scan script (use the --budget value from arguments, or 100 if not specified):

node ${CLAUDE_PLUGIN_ROOT}/scripts/graph.js scan --dir . --budget {budget}

The script returns JSON wrapped in {status, data, error} envelope format. Parse the data field for results.

Step 2: The script returns a JSON manifest with files[]. Each file has path, language, type, and size. For each file with type of source, manifest, or config: read the file and add:

• description: max 12 words describing its purpose
• exports: array of function/class/constant names exported by the file (empty array for non-source files)

For files you cannot or do not read: leave description and exports absent.

Step 3: Write the enriched manifest to docs/professor/architecture/.build/scan-manifest.json. Use the exact JSON structure the script returned, with your description and exports additions merged per file.

Step 4: Output ONLY this JSON (no prose, no markdown): {"wrote": "scan-manifest.json", "files_found": <total_files from script output>}