cqs 1.15.2

Code intelligence and RAG for AI agents. Semantic search, call graphs, impact analysis, type dependencies, and smart context assembly — in single tool calls. 52 languages + L5X/L5K PLC exports, 90.9% Recall@1 (BGE-large), 0.949 MRR (296 queries). Local ML, GPU-accelerated.
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
# Contributing to cqs

Thank you for your interest in contributing to cqs!

## Development Setup

**Requires Rust 1.93+** (check with `rustc --version`)

1. Clone the repository:
   ```bash
   git clone https://github.com/jamie8johnson/cqs
   cd cqs
   ```

2. Build:
   ```bash
   cargo build                        # CPU-only
   cargo build --features gpu-index   # with GPU acceleration (requires CUDA)
   ```

3. Run tests:
   ```bash
   cargo test                         # CPU-only
   cargo test --features gpu-index    # with GPU acceleration
   ```

4. Initialize and index (for manual testing):
   ```bash
   cargo run -- init
   cargo run -- index
   cargo run -- "your search query"
   ```

5. Set up pre-commit hook (recommended):
   ```bash
   git config core.hooksPath .githooks
   ```
   This runs `cargo fmt --check` before each commit.

## Code Style

- Run `cargo fmt` before committing
- No clippy warnings: `cargo clippy -- -D warnings`
- Add tests for new features
- Follow existing code patterns

### `_with_*` Function Naming Convention

Functions that accept pre-loaded resources use a `_with_<resource>` suffix:

| Suffix | Meaning | Example |
|--------|---------|---------|
| `_with_graph` | Pre-loaded call graph | `gather_with_graph()` |
| `_with_options` | Config struct parameter | `scout_with_options()` |
| `_with_embedding` | Pre-computed embedding | `suggest_placement_with_embedding()` |
| `_with_resources` | Pre-loaded embedder + graph | `task_with_resources()` |

Rules:
- The base function loads its own resources. The `_with_*` variant accepts them.
- Don't stack suffixes (`_with_graph_depth`). Add parameters to the existing `_with_*` function instead.
- If the `_with_*` variant has no external callers, fold it into the base function.

### JSON Output Field Naming Conventions

All `--json` output uses consistent field names across commands:

| Field | Not | Why |
|-------|-----|-----|
| `line_start` / `line_end` | `line`, `lines` | Separate scalars, not an array or ambiguous singular |
| `name` | `function`, `identifier` | Works for structs, enums, traits, not just functions |
| `score` | `similarity` | Generic — covers RRF, cosine, and risk scores |
| `file` | `origin`, `path` | Matches user mental model; `origin` is too abstract |

Rules:
- **snake_case** for all field names — no camelCase, no kebab-case.
- All output structs use `#[derive(serde::Serialize)]` with serde's default snake_case renaming. Do not use `#[serde(rename = "...")]` unless matching an external schema.
- Use `#[serde(skip_serializing_if = "Option::is_none")]` for optional fields so absent data is omitted (not `null`).
- When adding `--json` to a new command, follow existing output structs (e.g., `ChunkSummary`, `CallerDetail`, `ExplainOutput`) rather than inventing new field names.

## Pull Request Process

1. Fork the repository and create a feature branch
2. Make your changes
3. Ensure all checks pass:
   ```bash
   cargo test --features gpu-index
   cargo clippy --features gpu-index -- -D warnings
   cargo fmt --check
   ```
4. Update documentation if needed (README, CLAUDE.md)
5. Submit PR against `main`

## What to Contribute

### Good First Issues

- Look for issues labeled `good-first-issue`
- Documentation improvements
- Test coverage improvements

### Feature Ideas

- Additional language support (see `src/language/` for current list — 52 languages + L5X/L5K PLC exports)
- Non-CUDA GPU support (ROCm for AMD, Metal for Apple Silicon)
- VS Code extension
- Performance improvements
- CLI enhancements

### Bug Reports

When reporting bugs, please include:
- cqs version (`cqs --version`)
- OS and architecture
- Steps to reproduce
- Expected vs actual behavior

## Architecture Overview

```
src/
  cli/          - Command-line interface (clap)
    mod.rs      - Top-level CLI module, re-exports
    definitions.rs - Clap argument definitions and command enum
    dispatch.rs - Command dispatch (match on command, call handlers)
    commands/   - Command implementations (organized by category)
      mod.rs      - Top-level re-exports
      resolve.rs  - Target resolution (function name → chunk)
      search/     - query, gather, similar, related, where_cmd, scout, onboard, neighbors
      graph/      - callers, deps, explain, impact, impact_diff, test_map, trace
      review/     - diff_review, ci, dead, health, suggest, affected
      index/      - build, gc, stale, stats
      io/         - blame, brief, context, diff, drift, notes, read, reconstruct
      infra/      - audit_mode, convert, doctor, init, project, reference, telemetry_cmd
      train/      - export_model, plan, task, train_data, train_pairs
    chat.rs     - Interactive REPL (wraps batch mode with rustyline)
    batch/      - Batch mode: persistent Store + Embedder, stdin commands, JSONL output, pipeline syntax
      mod.rs      - BatchContext, vector index builder, main loop
      commands.rs - BatchInput/BatchCmd parsing, dispatch router
      handlers/ - Handler functions (one per command)
        mod.rs, analysis.rs, graph.rs, info.rs, misc.rs, search.rs
      pipeline.rs - Pipeline execution (pipe chaining via `|`)
      types.rs    - Output types (ChunkOutput, normalize_path)
    args.rs     - Shared CLI/batch arg structs via #[command(flatten)]
    config.rs   - Configuration file loading
    display.rs  - Output formatting, result display
    enrichment.rs - Enrichment pass (extracted from pipeline.rs)
    files.rs    - File enumeration, lock files, path utilities
    pipeline/   - Multi-threaded indexing pipeline
      mod.rs, embedding.rs, parsing.rs, types.rs, upsert.rs, windowing.rs
    signal.rs   - Signal handling (Ctrl+C)
    staleness.rs - Proactive staleness warnings for search results
    telemetry.rs - Optional command usage logging (CQS_TELEMETRY=1)
    store.rs    - Store opening utilities, CommandContext, vector index building
    watch.rs    - File watcher for incremental reindexing
  language/     - Tree-sitter language support
    mod.rs      - Language enum, LanguageRegistry, LanguageDef, ChunkType
    rust.rs, python.rs, typescript.rs, javascript.rs, go.rs, c.rs, cpp.rs, java.rs, csharp.rs, fsharp.rs, powershell.rs, scala.rs, ruby.rs, bash.rs, hcl.rs, kotlin.rs, swift.rs, objc.rs, sql.rs, protobuf.rs, graphql.rs, php.rs, lua.rs, zig.rs, r.rs, yaml.rs, toml_lang.rs, elixir.rs, erlang.rs, gleam.rs, haskell.rs, julia.rs, ocaml.rs, css.rs, perl.rs, html.rs, json.rs, xml.rs, ini.rs, nix.rs, make.rs, latex.rs, solidity.rs, cuda.rs, glsl.rs, svelte.rs, razor.rs, vbnet.rs, vue.rs, aspx.rs, markdown.rs, structured_text.rs
  test_helpers.rs - Shared test fixtures module
  store/        - SQLite storage layer (Schema v16, WAL mode)
    mod.rs      - Store struct, open/init, FTS5
    metadata.rs - Chunk metadata queries, file-level operations
    search.rs   - RRF fusion, search_filtered, search_unified_with_index
    chunks/     - Chunk storage and retrieval
      mod.rs, crud.rs, staleness.rs, embeddings.rs, query.rs, async_helpers.rs
    notes.rs    - Note CRUD, note_embeddings(), brute-force search
    calls/      - Call graph storage and queries
      mod.rs, crud.rs, dead_code.rs, query.rs, related.rs, test_map.rs
    types.rs    - Type edge storage and queries
    helpers/    - Types, embedding conversion, scoring, SQL utilities
      mod.rs, embeddings.rs, error.rs, rows.rs, scoring.rs, search_filter.rs, sql.rs, types.rs
    migrations.rs - Schema migration framework
  parser/       - Code parsing (tree-sitter + custom parsers, delegates to language/ registry)
    mod.rs      - Parser struct, parse_file(), parse_file_all(), supported_extensions()
    types.rs    - Chunk (incl. parent_type_name), CallSite, FunctionCalls, TypeRef, ParserError
    chunk.rs    - Chunk extraction, signatures, doc comments, parent type extraction
    calls.rs    - Call graph extraction, callee filtering
    injection.rs - Multi-grammar injection (HTML→JS/CSS via set_included_ranges)
    aspx.rs     - ASP.NET Web Forms (.aspx/.ascx/.asmx) custom parser
    l5x.rs      - Rockwell PLC exports (L5X XML + L5K ASCII) → Structured Text extraction
    markdown/   - Heading-based markdown parser
      mod.rs, headings.rs, code_blocks.rs, tables.rs
  embedder/      - ONNX embedding models (configurable: BGE-large-en-v1.5 default, E5-base preset, custom ONNX)
    mod.rs      - Embedder struct, embed(), batch embedding, runtime dimension detection
    models.rs   - ModelConfig struct, built-in presets (e5-base, bge-large), resolution logic, EmbeddingConfig
    provider.rs - ORT execution provider selection (CUDA/TensorRT/CPU)
  reranker.rs   - Cross-encoder re-ranking (ms-marco-MiniLM-L-6-v2)
  search/       - Search algorithms, name matching, HNSW-guided search
    mod.rs      - search_filtered(), search_unified_with_index(), hybrid RRF
    scoring/    - ScoringConfig, score normalization, RRF fusion constants
      mod.rs, candidate.rs, config.rs, filter.rs, name_match.rs, note_boost.rs
    query.rs    - Query parsing, filter extraction
    synonyms.rs - Query synonym expansion
  math.rs       - Vector math utilities (cosine similarity, SIMD)
  hnsw/         - HNSW index with batched build, atomic writes
    mod.rs      - HnswIndex, LoadedHnsw (self_cell), HnswError, VectorIndex impl
    build.rs    - build(), build_batched() construction
    search.rs   - Nearest-neighbor search
    persist.rs  - save(), load(), checksum verification
    safety.rs   - Send/Sync and loaded-index safety tests
  convert/      - Document-to-Markdown conversion (optional, "convert" feature)
    mod.rs      - ConvertOptions, convert_path(), format detection
    html.rs     - HTML → Markdown via fast_html2md
    pdf.rs      - PDF → Markdown via Python pymupdf4llm (shell out)
    chm.rs      - CHM → 7z extract → HTML → Markdown
    naming.rs   - Title extraction, kebab-case filename generation
    cleaning.rs - Extensible tag-based cleaning rules (7 rules)
    webhelp.rs  - Web help site detection and multi-page merge
  cagra.rs      - GPU-accelerated CAGRA index (optional)
  nl/           - NL description generation, JSDoc parsing
    mod.rs      - Core NL generation, type-aware embeddings, call context
    fts.rs      - FTS5 normalization, tokenization
    fields.rs   - Field/keyword extraction from code bodies
    markdown.rs - Markdown-specific NL generation
  note.rs       - Developer notes with sentiment, rewrite_notes_file()
  diff.rs       - Semantic diff between indexed snapshots
  drift.rs      - Drift detection (semantic change magnitude between snapshots)
  reference.rs  - Multi-index: ReferenceIndex, load, search, merge
  gather.rs     - Smart context assembly (BFS call graph expansion)
  structural.rs - Structural pattern matching on code chunks
  project.rs    - Cross-project search registry
  audit.rs    - Audit mode persistence and duration parsing
  focused_read.rs - Focused read logic (extract type dependencies)
  impact/         - Impact analysis (callers + affected tests + diff-aware)
    mod.rs      - Public API, re-exports
    types.rs    - Impact types (CallerDetail, RiskScore, etc.)
    analysis.rs - suggest_tests, find_transitive_callers, extract_call_snippet_from_cache
    diff.rs     - analyze_diff_impact, map_hunks_to_functions
    bfs.rs      - Reverse BFS, reverse_bfs_multi_attributed, test_reachability
    format.rs   - JSON/Mermaid formatting
    hints.rs    - compute_hints, compute_hints_batch, compute_risk_batch, risk scoring
    test_map.rs - Shared test-map algorithm (reverse BFS from function to test chunks)
  related.rs      - Co-occurrence analysis (shared callers, callees, types)
  scout.rs        - Pre-investigation dashboard (search + callers/tests + staleness + notes)
  task.rs         - Single-call implementation brief (scout + gather + impact + placement + notes)
  onboard.rs      - Guided codebase tour (entry point + call chain + callers + types + tests)
  review.rs       - Diff review (impact-diff + notes + risk scoring)
  ci.rs           - CI pipeline (review + dead code + gate logic)
  where_to_add.rs - Placement suggestion (semantic search + pattern extraction)
  plan.rs         - Task planning with 11 task-type templates
  diff_parse.rs   - Unified diff parser for impact-diff
  health.rs     - Codebase quality snapshot (dead code, staleness, hotspots)
  suggest.rs    - Auto-suggest notes from code patterns
  config.rs     - Configuration file support
  index.rs      - VectorIndex trait (HNSW, CAGRA)
  llm/          - LLM summary generation, HyDE query predictions via Anthropic Batches API
    mod.rs, batch.rs (BatchPhase2, submit_batch_prebuilt), doc_comments.rs, hyde.rs, prompts.rs (build_contrastive_prompt), provider.rs (BatchProvider trait, BatchSubmitItem, LlmProvider), summary.rs (find_contrastive_neighbors)
  doc_writer/   - Doc comment generation and source file rewriting (SQ-8, optional "llm-summaries" feature)
    mod.rs      - DocCommentResult, module exports
    formats.rs  - Per-language doc comment formatting (prefix, position, wrapping)
    rewriter.rs - Source file rewriter: find insertion point, apply edits bottom-up, atomic write
  train_data/   - Fine-tuning training data generation from git history
    mod.rs      - TrainDataConfig, generate_training_data(), Triplet types
    bm25.rs     - BM25 index for hard negative mining
    checkpoint.rs - Resume support for long generation runs
    diff.rs     - Git diff parsing for function-level changes
    git.rs      - Git history traversal (log, show, diff-tree)
    query.rs    - Query normalization for training pairs
  lib.rs        - Public API
.claude/
  skills/       - Claude Code skills (auto-discovered)
    groom-notes/  - Interactive note review and cleanup
    update-tears/ - Session state capture for context persistence
    release/      - Version bump, changelog, publish workflow
    audit/        - 14-category code audit with parallel agents
    red-team/     - Adversarial security audit (attacker mindset, PoC-required)
    pr/           - WSL-safe PR creation
    cqs-bootstrap/ - New project setup with tears infrastructure
    cqs/          - Unified CLI dispatcher (search, graph, quality, notes, infrastructure)
    reindex/      - Rebuild index with before/after stats
    docs-review/  - Check project docs for staleness
    migrate/      - Schema version upgrades
    troubleshoot/ - Diagnose common cqs issues
    cqs-batch/    - Batch mode with pipeline syntax
    cqs-plan/     - Task planning with templates
    before-edit/  - Pre-edit workflow: snapshot state before changes
    investigate/  - Investigation workflow: structured code exploration
    check-my-work/ - Post-implementation verification checklist
    cqs-verify/   - Exercise all command categories, catch regressions
```

**Key design notes:**
- Configurable embeddings (BGE-large 1024-dim default, E5-base 768-dim preset, custom ONNX)
- HNSW index is chunk-only; notes use brute-force SQLite search (always fresh)
- Streaming HNSW build via `build_batched()` for memory efficiency
- Large chunks split by windowing (480 tokens, 64 overlap); notes capped at 10k entries
- Schema migrations allow upgrading indexes without full rebuild
- Skills in `.claude/skills/*/SKILL.md` are auto-discovered by Claude Code

## Adding a New CLI Command

Checklist for every new command:

1. **Implementation** — `src/cli/commands/<category>/<name>.rs` with the core logic (pick category: search/, graph/, review/, index/, io/, infra/, train/)
2. **Category mod.rs** — add `mod <name>;` + `pub(crate) use <name>::*;` in `src/cli/commands/<category>/mod.rs`
3. **CLI definition** — `Commands` enum variant in `src/cli/definitions.rs` with clap args
4. **Dispatch** — match arm in `src/cli/dispatch.rs`
5. **`--json` support** — serde serialization for programmatic output
6. **Tracing** — `tracing::info_span!` at entry, `tracing::warn!` on error fallback
7. **Error handling** — `Result` propagation, no bare `.unwrap_or_default()` in production
8. **Tests** — happy path + empty input + error path + edge cases
9. **CLAUDE.md** — add to the command reference section
10. **Skills** — add to `.claude/skills/cqs/SKILL.md` and `.claude/skills/cqs-bootstrap/SKILL.md`
11. **CHANGELOG** — entry in the next release section

Pattern to follow: look at `src/cli/commands/io/blame.rs` or `src/cli/commands/review/dead.rs` for a minimal example.

## Adding Injection Rules (Multi-Grammar)

Files like HTML contain embedded languages (`<script>` → JS, `<style>` → CSS). cqs handles this via injection rules on `LanguageDef`.

**To add injection rules for a new host language:**

1. Define `InjectionRule` entries in the language's `LanguageDef` (`src/language/<lang>.rs`):
   ```rust
   injections: &[
       InjectionRule {
           container_kind: "script_element",  // outer tree node kind
           content_kind: "raw_text",          // child node with embedded content
           target_language: "javascript",     // must match a Language variant name
           detect_language: Some(detect_fn),  // optional: inspect attributes for lang override
       },
   ],
   ```

2. `container_kind` / `content_kind` must match the host grammar's node kinds (inspect with `tree-sitter parse`).

3. `target_language` must be a valid `Language` name with a grammar (validated at runtime in `find_injection_ranges`).

4. `detect_language` receives the container node and source — return `Some("typescript")` to override the default, `Some("_skip")` to skip the container entirely, or `None` for the default.

5. Injection is single-level only. Inner languages are not re-scanned for their own injections.

6. The two-phase flow in `parse_file` and `parse_file_relationships` automatically handles injection when `injections` is non-empty. No changes needed outside the language definition.

**Key files:** `src/language/mod.rs` (InjectionRule struct), `src/parser/injection.rs` (parsing logic), `src/language/html.rs` (reference implementation).

## Adding a New Language

Adding a language is a data-entry task, not a coding task. The `LanguageDef` system handles everything — you fill in fields.

### Prerequisites

- A tree-sitter grammar published on crates.io (search `tree-sitter-<lang>`)
- A sample source file to test with
- `tree-sitter parse sample.ext` to see node types (install: `cargo install tree-sitter-cli`)

### Steps

**1. Add the dependency to `Cargo.toml`:**

```toml
tree-sitter-dart = { version = "0.X", optional = true }
```

And the feature flag:
```toml
lang-dart = ["dep:tree-sitter-dart"]
```

Add `"lang-dart"` to the `default` and `lang-all` feature lists.

**2. Create `src/language/dart.rs`:**

Copy `src/language/bash.rs` as your starting template — it's the simplest language file (~65 lines). Then fill in:

```rust
//! Dart language definition

use super::{FieldStyle, LanguageDef, SignatureStyle};

// === STEP A: Write the chunk query ===
// Run `tree-sitter parse sample.dart` and look for function-like nodes.
// Common patterns: function_declaration, method_declaration, class_declaration
const CHUNK_QUERY: &str = r#"
(function_declaration
  name: (identifier) @name) @function

(method_declaration
  name: (identifier) @name) @function

(class_declaration
  name: (identifier) @name) @class
"#;

// === STEP B: Write the call query ===
// Look for call-like nodes in the AST dump.
const CALL_QUERY: &str = r#"
(call_expression
  function: (identifier) @callee)

(call_expression
  function: (selector) @callee)
"#;

// === STEP C: Fill in the rest (data entry) ===
const DOC_NODES: &[&str] = &["comment", "documentation_comment"];

const STOPWORDS: &[&str] = &[
    "if", "else", "for", "while", "do", "return", "class", "extends",
    "implements", "import", "void", "var", "final", "const", "static",
    "this", "super", "new", "null", "true", "false", "async", "await",
];

const COMMON_TYPES: &[&str] = &[
    "String", "int", "double", "bool", "List", "Map", "Set", "Future",
    "Stream", "void", "dynamic", "Object", "Iterable", "Function",
];

static DEFINITION: LanguageDef = LanguageDef {
    name: "dart",
    grammar: Some(|| tree_sitter_dart::LANGUAGE.into()),
    extensions: &["dart"],
    chunk_query: CHUNK_QUERY,
    call_query: Some(CALL_QUERY),
    signature_style: SignatureStyle::UntilBrace,
    doc_nodes: DOC_NODES,
    method_node_kinds: &["method_declaration"],
    method_containers: &["class_body"],
    stopwords: STOPWORDS,
    extract_return_nl: |sig| {
        // Dart: ReturnType functionName(params) { ... }
        // Type is before the function name
        None // Start simple, add later
    },
    test_file_suggestion: None,
    test_name_suggestion: None,
    type_query: None,       // Add later for type edges
    common_types: COMMON_TYPES,
    container_body_kinds: &["class_body"],
    extract_container_name: None,
    extract_qualified_method: None,
    post_process_chunk: None,
    test_markers: &["@test", "test("],
    test_path_patterns: &["%_test.dart", "%/test/%"],
    structural_matchers: None,
    entry_point_names: &["main"],
    trait_method_names: &[],
    injections: &[],
    doc_format: "default",
    doc_convention: "Use /// for documentation comments. Follow Effective Dart documentation guidelines.",
    field_style: FieldStyle::NameFirst {
        separators: ":",
        strip_prefixes: "final late var static const",
    },
};

pub fn definition() -> &'static LanguageDef {
    &DEFINITION
}
```

**3. Register in `src/language/mod.rs`:**

Add one line to `define_languages!`:
```rust
Dart => "dart", feature = "lang-dart", module = dart;
```

**4. Write tests:**

Add a `#[cfg(test)] mod tests` section in your file. Minimum 3 tests:
- Parse a function → verify name and ChunkType::Function
- Parse a class → verify name and ChunkType::Class
- Parse function calls → verify callee names extracted

See `src/language/bash.rs` tests for the pattern.

**5. Build and test:**

```bash
cargo test --features gpu-index,lang-dart -- dart
```

### How to discover node types

Run `tree-sitter parse` on a sample file:

```bash
tree-sitter parse sample.dart 2>/dev/null | head -50
```

Output shows the AST. Look for:
- **Function nodes**: usually `function_declaration`, `method_declaration`, `function_expression`
- **Class nodes**: `class_declaration`, `interface_declaration`, `enum_declaration`
- **Call nodes**: `call_expression`, `method_invocation`
- **Name fields**: `name:` or `(identifier)`

The chunk query captures `@name` (the function/class name) and `@function` / `@class` / `@property` etc. (the full node for content extraction).

### Fields Reference

Most fields have sensible defaults (`None`, `&[]`, empty string). The important ones:

| Field | Required? | How to fill |
|-------|-----------|-------------|
| `grammar` | Yes | `Some(\|\| tree_sitter_<lang>::LANGUAGE.into())` |
| `extensions` | Yes | File extensions without dot |
| `chunk_query` | Yes | Tree-sitter S-expression query |
| `call_query` | Recommended | Tree-sitter query for function calls |
| `signature_style` | Yes | `UntilBrace` for C-like, `UntilNewline` for Python-like |
| `doc_nodes` | Recommended | Node kinds that contain doc comments |
| `stopwords` | Recommended | Language keywords to filter from NL |
| `common_types` | Recommended | Stdlib types to exclude from type edges |
| `field_style` | Recommended | `NameFirst`/`TypeFirst`/`None` for struct field extraction |
| Everything else | Optional | `None`, `&[]`, or `""` — add later as needed |

### Ecosystem updates (after the language works)

- Add `"lang-dart"` to the default features list in `Cargo.toml`
- Add to `CLAUDE.md` agent instructions (key commands block in agent prompts)
- Add to `README.md` language count
- Update `CHANGELOG.md`

## Questions?

Open an issue for questions or discussions.