NoWreck v0.3.0 — JavaScript Support 🎉

Deterministic verification for AI code changes — now for JavaScript too.

What's new in v0.3.0

The headline feature is JavaScript support — NoWreck can now scan, parse, and verify JavaScript files alongside Python files in the same repository. This is the first time the pipeline has been extended beyond Python, and it's designed so that adding future languages follows the same pattern.

JavaScript scanning via Tree-sitter

JavaScript files (.js) are now parsed using Tree-sitter with the tree-sitter-javascript grammar, producing the same Symbol / SymbolType data shapes as the Python parser. The rest of the pipeline — change detection, claim verification, and reporting — never knows or cares which language produced the data.

What the JS scanner captures:

Pattern	Example	Status
Function declarations	`function greet() {}`	✅
Arrow functions (assigned)	`const greet = () => {}`	✅
Classes	`class Calculator {}`	✅
Class methods	`add(a, b) { ... }`	✅
`export function`	`export function greet() {}`	✅
`export class`	`export class Calculator {}`	✅
`export const` (arrow)	`export const greet = () => {}`	✅
Function calls	`greet()` in function bodies	✅ (simple calls only)

Mixed-language repos

NoWreck handles repositories with both .py and .js files in a single scan. The pre/post summaries show separate file counts for each language:

Scan Summary
────────────
Python:    5 files (3 success, 0 failed)
JavaScript: 3 files (3 success, 0 failed)
Symbols:   22 total (12 functions, 4 classes, 6 methods)

Same 7 claim types — unchanged

No new claim types were added. The same 7 deterministic claims work across both languages:

ADD_FUNCTION / REMOVE_FUNCTION
ADD_CLASS / REMOVE_CLASS
FILE_CREATED / FILE_DELETED
CALLS_FUNCTION

A claim about a Python function and a claim about a JavaScript function are verified identically — the verifier doesn't need to know which language produced the data.

Lazy dependency loading

The tree-sitter-javascript dependency is loaded lazily — it's only imported when a JavaScript file is actually scanned. This means:

Pure-Python repos work without tree-sitter-javascript installed
No cascading import failures in the Python-only test suite
All 388 existing pytest tests pass without the JS grammar

Full changelog

Added

New module: nowreck/scanner/javascript_scanner.py — Tree-sitter-based JavaScript parser with lazy grammar loading
- scan_js_file(path) — parses a .js file and returns list[Symbol]
- scan_js_calls(source_code, symbols) — extracts call_expression nodes and returns list[DetectedChange] for CALL_DETECTED
- _get_js_language() — lazy-loads tree-sitter-javascript with double-checked caching
SymbolIndex.js_symbols property — returns only JavaScript symbols (those without an ast module origin)
SymbolIndex.by_name(name) — language-agnostic lookup (finds symbols across both languages)
RepositoryScanner extended — discovers .js files, routes them to javascript_scanner, populates ScanResult.js_files
ChangeDetector.detect() extended — extracts JS call_expression nodes alongside Python ast.Call walking
TerminalReporter extended — shows Python vs JS file counts and language-specific symbol breakdowns in scan summaries
_strip_markdown_fence() in claims/parser.py — handles ```json...``` from model output

Changed

Bumped version from v0.2.0 → v0.3.0
Updated pyproject.toml — added tree-sitter-javascript>=0.25 dependency
Updated .gitignore — added _binary_test.bin under test artifacts
Updated README.md — JS integration documented in scan stage, limitations, roadmap, troubleshooting, and tips
Updated use.md — version references, JSON schema examples
Fixed pre-existing ruff issues in tests/test_picker.py and tests/test_picker_integration.py (unused imports, line lengths)
Fixed pre-existing type errors in nowreck/picker.py (missing questionary stub)
Fixed ordering-sensitive flaky test (test_empty_prompt_flow) in test_picker_integration.py

Known limitations (v0.3.0)

Generator functions (function*) — not captured (deferred)
export default — not captured (deferred)
IIFEs — not captured (deferred)
TypeScript — not yet supported (separate scope, deferred)
Attribute calls (e.g., console.log(), obj.method()) — excluded from CALLS_FUNCTION detection; only simple name() calls are tracked
All existing Python limitations still apply (no dynamic behavior, no cross-file resolution beyond direct name matching, no semantic analysis)

Test statistics

Suite	Tests	Result
Python unit tests (pytest)	388	✅ 388/388 pass
JS multi-round (repeatability, stress, chaos)	80	✅ 80/80 pass
JS comprehensive (core, edge cases, negatives, error handling)	78	✅ 78/78 pass
Milestone 1 — pure Python repo	10	✅ Deterministic (3x)
Milestone 1 — pure JS repo	10	✅ Deterministic (3x)
Milestone 1 — mixed repo	10	✅ Deterministic (3x)
Full pipeline determinism (mixed repo, 5 runs)	1	✅ Identical every run
Phase 4a end-to-end demo (14 hand-written claims)	14	✅ Pipeline clean
Phase 4d live-model hallucination-catch (real API)	1	✅ Hallucination caught
Total	~590	✅ All pass
Linting (ruff)	—	✅ 0 issues
Type checking (basedpyright)	—	✅ 0 errors, 0 warnings, 0 notes

How it works

┌─────────────────────────────────────────────────────────┐
│                   Prompt mode                           │
│                                                         │
│  Your prompt ──► AI model ──► diff + claims             │
│                                   │                     │
│                                   ▼                     │
│  Pre-scan ──► Symbol index ──► Change Detector          │
│  Post-scan ──► Symbol index ────────┘                   │
│                                         │               │
│  Claims ──► Claim Verifier ◄────────────┘               │
│                  │     pure comparison — no AI judgment  │
│                  ▼                                      │
│          Verification Report                            │
│   ✓ CONFIRMED  ✗ CONTRADICTED  ? UNVERIFIABLE           │
└─────────────────────────────────────────────────────────┘

Scan stage (now with dual-language support)

Discover — recursively finds .py and .js files in both snapshots
Parse — routes each file to the correct parser:
- .py → Python's built-in ast module
- .js → Tree-sitter with tree-sitter-javascript grammar
Index — builds a unified SymbolIndex from both languages

Both parsers produce identical Symbol / SymbolType data shapes (using LanguageAdapter patterns anticipated in earlier architecture). The rest of the pipeline is language-agnostic.

Quick start with JavaScript

# Create a JS test repo
mkdir -p /tmp/js-app/pre /tmp/js-app/post

cat > /tmp/js-app/pre/greeter.js << 'EOF'
function greet(name) {
  return "Hello, " + name + "!";
}
EOF

cat > /tmp/js-app/post/greeter.js << 'EOF'
function greet(name) {
  return "Hello, " + name + "!";
}

const farewell = (name) => "Goodbye, " + name + "!";
EOF

# Detect changes
nowreck fix --pre /tmp/js-app/pre --post /tmp/js-app/post

# Or with claims that include a hallucinated call
nowreck fix \
  --pre /tmp/js-app/pre \
  --post /tmp/js-app/post \
  --claims '{
    "claims": [
      {
        "type": "ADD_FUNCTION",
        "symbol_name": "farewell",
        "file_path": "greeter.js",
        "confidence": 0.99,
        "explanation": "Added farewell arrow function."
      },
      {
        "type": "CALLS_FUNCTION",
        "symbol_name": "farewell",
        "file_path": "greeter.js",
        "caller_name": "farewell",
        "called_name": "notify",
        "confidence": 0.85,
        "explanation": "farewell calls notify to send the message."
      }
    ]
  }'

# Expected: ADD_FUNCTION CONFIRMED, CALLS_FUNCTION CONTRADICTED
# (farewell exists, but it doesn't call notify)

Mixed Python + JavaScript

nowreck fix --pre ./before --post ./after
# → Shows Python: 3 files, JavaScript: 2 files in scan summary

Design decisions

Why Tree-sitter?

JavaScript has no equivalent of Python's built-in ast module, so an external parser was required. Tree-sitter was chosen because:

Mature JS grammar — well-maintained tree-sitter-javascript with broad coverage
Reusable pattern — same parser family would be used for Go, Rust, etc.
Concrete syntax trees — provides real source positions for accurate claim-to-diff line mapping

Why lazy imports?

The tree-sitter-javascript grammar is imported lazily (only when a .js file is actually scanned). This ensures:

The Python-only test suite doesn't cascade-fail on import
Pure-Python repos work without the JS grammar installed
The dependency is a hard install requirement (in pyproject.toml) but a soft runtime requirement

Why no new claim types?

The existing 7 claim types (ADD_FUNCTION, REMOVE_FUNCTION, ADD_CLASS, REMOVE_CLASS, FILE_CREATED, FILE_DELETED, CALLS_FUNCTION) are language-agnostic by design. Adding JS-specific claim types (e.g., around export/import patterns) is a deliberate future-scope decision, not something to add mid-build because it seems easy.

Architecture summary

Component	v0.2.0 (Python only)	v0.3.0 (Python + JS)
Scanner	`ast` only	`ast` + Tree-sitter
Symbol index	Python symbols only	Unified index (both languages)
Change detector	`ast.Call` walking	`ast.Call` + Tree-sitter `call_expression`
Claim verifier	Language-agnostic	Unchanged — still language-agnostic
Reporter	Python-only summaries	Python + JS breakdowns
CLI interface	—	Unchanged — fully backwards compatible

Credits

Built using Tree-sitter and the tree-sitter-javascript grammar. All existing Python infrastructure remains unchanged — this was purely additive.

NoWreck v0.3.0 — July 2026

Navigation Menu

NoWreck v0.3.0 — JavaScript Support 🎉

Choose a tag to compare

Sorry, something went wrong.

Sorry, something went wrong.

Uh oh!

No results found

NoWreck v0.3.0 — JavaScript Support 🎉

What's new in v0.3.0

JavaScript scanning via Tree-sitter

Mixed-language repos

Same 7 claim types — unchanged

Lazy dependency loading

Full changelog

Added

Changed

Known limitations (v0.3.0)

Test statistics

How it works

Scan stage (now with dual-language support)

Quick start with JavaScript

Mixed Python + JavaScript

Design decisions

Why Tree-sitter?

Why lazy imports?

Why no new claim types?

Architecture summary

Credits

Uh oh!