feat: multi-language tree-sitter extractor architecture (10 languages)

[Lum1104]

Summary

• Decouple AST extraction from TS/JS-specific node types — introduced LanguageExtractor interface with per-language extractor classes, refactored TreeSitterPlugin (746→297 lines) to dispatch via registered extractors
• Add tree-sitter structural analysis for 10 languages — TypeScript, JavaScript, Python, Go, Rust, Java, Ruby, PHP, C/C++, C# each get a dedicated extractor with full extractStructure() + extractCallGraph() support (239 new tests)
• Replace LLM-generated regex scripts with deterministic extraction — bundled extract-structure.mjs uses PluginRegistry (tree-sitter + non-code parsers) so the file-analyzer agent no longer writes throwaway scripts from scratch each run
• Fix GraphBuilder duplicate language mapping — replaced hardcoded EXTENSION_LANGUAGE map with LanguageRegistry delegation

Architecture

TreeSitterPlugin
  ├── registerExtractor(extractor)
  ├── getExtractor(langKey) → LanguageExtractor
  ├── analyzeFile() → delegates to extractor.extractStructure()
  └── extractCallGraph() → delegates to extractor.extractCallGraph()

extractors/
  ├── types.ts              # LanguageExtractor interface
  ├── base-extractor.ts     # Shared AST utilities
  ├── typescript-extractor.ts
  ├── python-extractor.ts
  ├── go-extractor.ts
  ├── rust-extractor.ts
  ├── java-extractor.ts
  ├── ruby-extractor.ts
  ├── php-extractor.ts
  ├── cpp-extractor.ts
  ├── csharp-extractor.ts
  └── index.ts              # builtinExtractors array

Changes by area

Core extractors (new)

• LanguageExtractor interface + shared traverse/findChild/getStringValue utilities
• 9 language-specific extractors (TS/JS shares one), each handling language-specific AST node types for functions, classes, imports, exports, and call graphs
• builtinExtractors array auto-registered in TreeSitterPlugin constructor

TreeSitterPlugin refactor

• Removed all TS/JS-specific extraction methods (moved to TypeScriptExtractor)
• Added extractor dispatch: registerExtractor(), getExtractor() with tsx→typescript mapping
• Constructor accepts optional extractors param, defaults to builtinExtractors

Plugin infrastructure

• PluginRegistry.extractCallGraph() — new delegation method
• DEFAULT_PLUGIN_CONFIG now derives languages dynamically from builtinLanguageConfigs.filter(c => c.treeSitter)
• 8 new tree-sitter grammar deps + treeSitter configs for Python, Go, Rust, Java, Ruby, PHP, C++, C#

Agent pipeline

• extract-structure.mjs — bundled script using PluginRegistry for deterministic extraction
• file-analyzer.md Phase 1 rewritten: executes bundled script instead of LLM-generated regex (−213 lines)
• SKILL.md passes <SKILL_DIR> to file-analyzer dispatch

Test plan

• All 693 tests pass (651 core + 42 skill)
• 239 new extractor tests across 8 test files (Python: 31, Go: 25, Rust: 30, Java: 25, Ruby: 32, PHP: 27, C++: 28, C#: 27)
• All 426 pre-existing tests pass unchanged (zero behavior change in TS/JS extraction)
• Core and skill packages build cleanly (tsc, zero errors)
• extract-structure.mjs tested locally against TS, Markdown, and JSON files
• Run /understand --full on a multi-language project to verify end-to-end pipeline

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6cba6de658

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Restore Lua language detection fallback

Routing detection entirely through LanguageRegistry here drops support for extensions that were only in the old map, notably .lua. Because builtinLanguageConfigs has no Lua config, addFile("*.lua", ...) now resolves to "unknown" and project.languages no longer includes "lua", which regresses graph metadata and downstream summaries for Lua repositories compared with the previous behavior.

Useful? React with 👍 / 👎.