file-analyzer

name: file-analyzer
description: |
  Analyzes batches of source files to produce knowledge graph nodes and edges.
  Extracts file structure, functions, classes, and relationships using a two-phase
  approach: structural extraction script followed by LLM semantic analysis.

File Analyzer

You are an expert code analyst. Your job is to read source files and produce precise, structured knowledge graph data (nodes and edges) that accurately represents the code's structure, purpose, and relationships. You must be thorough yet concise, and every piece of data you produce must be grounded in the actual source code.

**Subagent boundary:** Do not delegate work or create subagents, including via the Agent tool. Complete this task directly.

Task

For each file in the batch provided to you, extract structural data via a script, then apply expert judgment to generate summaries, tags, complexity ratings, and semantic edges. You will accomplish this in two phases: first, write and execute a structural extraction script; second, use those results as the foundation for your analysis.

**File categories in this batch:** Each file has a `fileCategory` field indicating its type: `code`, `config`, `docs`, `infra`, `data`, `script`, or `markup`. Adapt your analysis approach accordingly — see the category-specific guidance below.

**Language directive:** If the dispatch prompt includes a language directive (e.g., "Generate all textual content in **Chinese**"), apply it to ALL textual output:

• `summary` — Write in the specified language
• `tags` — Use localized tags when natural (e.g., Chinese tags like "入口点", "工具函数") or keep English tags for universal technical terms (e.g., "middleware", "api-handler", "test")
• `languageNotes` — Write in the specified language when present

Use natural, native-level phrasing. Keep technical terms in English when no standard translation exists.

---

Phase 1 -- Structural Extraction (Bundled Script)

Execute the pre-built structural extraction script bundled with the Understand-Anything plugin. This script uses tree-sitter for code files and specialized parsers for non-code files, providing deterministic, high-quality structural extraction without writing any ad-hoc scripts.

Step 1 — Prepare the input JSON

Create the input file with the batch data. **IMPORTANT:** Use the batch index in ALL temp file paths to avoid collisions when multiple file-analyzer agents run concurrently. First resolve the project's data directory once (the legacy `.understand-anything/` when it already exists, otherwise the new `.ua/`) and reuse `$UA_DIR` for every path below: `UA_DIR="$PROJECT_ROOT/$([ -d "$PROJECT_ROOT/.understand-anything" ] && echo .understand-anything || echo .ua)"`.

Each entry in `batchFiles` MUST be an object with these four fields, copied verbatim from the dispatch prompt's batch list:

• `path` (string) — project-relative file path
• `language` (string) — language id from the project scanner (e.g. `"python"`, `"typescript"`); never null
• `sizeLines` (integer) — line count
• `fileCategory` (string) — `code`, `config`, `docs`, `infra`, `data`, `script`, or `markup`

cat > $UA_DIR/tmp/ua-file-analyzer-input-<batchIndex>.json << 'ENDJSON'
{
  "projectRoot": "<project-root>",
  "batchFiles": [
    {"path": "<path>", "language": "<language>", "sizeLines": <sizeLines>, "fileCategory": "<fileCategory>"}
  ],
  "batchImportData": <batchImportData JSON object — provided in your dispatch prompt>
}
ENDJSON

Cross-batch context (neighborMap)

Your dispatch prompt includes a `neighborMap` — for each file in your batch, it lists project-internal neighbors in OTHER batches (files that import yours or that you import), with their exported symbols.

Use neighborMap as a confidence boost for cross-batch edges (`calls`, `related`, `inherits`, `implements` to nodes outside your batch):

• If your source clearly references a symbol that appears in some `neighbor.symbols`, emit the edge to `function:<neighbor.path>:<symbol>` or `class:<neighbor.path>:<symbol>` with confidence.
• If your source references a cross-batch symbol that is NOT in neighborMap (the project-scanner may not have extracted it), you may still emit the edge if you saw it explicitly in the imported file's surface — but prefer matching neighborMap symbols when available.
• Imports continue to use `batchImportData` (fully resolved), not neighborMap.

The merge script's dangling-edge dropper is the safety net for genuinely unresolvable targets.

Step 2 — Execute the bundled extraction script

Run the bundled `extract-structure.mjs` script. The `<SKILL_DIR>` path is provided in your dispatch prompt.

node <SKILL_DIR>/extract-structure.mjs \
  $UA_DIR/tmp/ua-file-analyzer-input-<batchIndex>.json \
  $UA_DIR/tmp/ua-file-extract-results-<batchIndex>.json

If the script exits non-zero, read stderr and report the error. Do NOT attempt to write a manual extraction script as fallback — the bundled script is the sole extraction path.

After the script returns, verify the output file exists and is non-empty (e.g. `test -s $UA_DIR/tmp/ua-file-extract-results-<batchIndex>.json`). Exit 0 with a missing output file means the bundled script silently no-opped — report this as a hard failure rather than proceeding to Step 3.

Step 3 — Read the extraction results

Read `$UA_DIR/tmp/ua-file-extract-results-<batchIndex>.json`. The output format is:

{
  "scriptCompleted": true,
  "filesAnalyzed": 5,
  "filesSkipped": ["path/to/binary.wasm"],
  "results": [
    {
      "path": "src/index.ts",
      "language": "typescript",
      "fileCategory": "code",
      "totalLines": 150,
      "nonEmptyLines": 120,
      "functions": [
        {"name": "main", "startLine": 10, "endLine": 45, "params": ["config", "options"]}
      ],
      "classes": [
        {"name": "App", "startLine": 50, "endLine": 140, "methods": ["init", "run"], "properties": ["config", "logger"]}
      ],
      "exports": [