feat: dev publish workflow, parser refactor, and quality-of-life improvements

[carlos-alm]

Summary

• Publish workflow consolidation: merge dev + stable publish into a single publish.yml workflow with workflow_dispatch inputs
• Parser refactor: split monolithic parser.js extractors into per-language files under src/extractors/
• codegraph stats command: new CLI command for graph health overview
• Native engine fixes: normalize import paths, throw on explicit --engine native when addon unavailable
• Registry improvements: TTL-based pruning for idle entries, auto-prune stale entries, skip temp dir registration, isolate CLI tests from real registry
• Embedding regression CI: real-model integration test with dedicated weekly workflow
• Worktree workflow hooks: guard-git.sh, track-edits.sh, rebuild-graph.sh for parallel session safety
• Windows compatibility: replace jq with node in hooks
• Competitive analysis: expanded from 21 to 135+ tools

Test plan

• Full test suite passes (385 passed, 43 skipped) on all platforms
• Lint clean
• Verify codegraph stats output on a real repo
• Verify publish.yml workflow_dispatch triggers correctly for dev and stable channels
• Confirm embedding regression workflow runs on schedule and relevant PR changes

[greptile-apps]

Greptile Summary

Consolidated dev+stable publish workflows into a single workflow_dispatch-triggered publish.yml that auto-computes versions and npm tags. Dev releases now publish automatically on every main commit with -dev.SHA suffix tagged as dev on npm, while stable releases extract version from GitHub release tags.

Refactored the 1917-line monolithic parser.js into a clean architecture with language-specific extractors in src/extractors/ (JavaScript, Python, Go, Rust, Java, C#, PHP, Ruby, HCL). All extraction logic moved to dedicated files with shared helpers, improving maintainability.

Added codegraph stats command that provides comprehensive graph health metrics: node/edge counts by kind, language distribution, cycle detection, top coupling hotspots, and embedding metadata.

Enhanced registry with TTL-based auto-pruning (30-day default) that removes idle entries, lastAccessedAt tracking updated on each resolution, and CODEGRAPH_REGISTRY_PATH env var support. Builder now skips auto-registration for temp directories.

Fixed native engine import path normalization issues by cleaning . and .. segments via PathBuf components in Rust and normalizing results in JS wrapper.

Implemented worktree workflow safety with three new hooks: guard-git.sh blocks dangerous git commands (add -A, reset, clean, stash) and validates commits against session edit log; track-edits.sh logs all Edit/Write operations; rebuild-graph.sh auto-rebuilds graph incrementally after source edits.

Replaced jq with node in all hooks for Windows compatibility.

Added embedding regression test with real ML model validation and dedicated weekly CI workflow.

All test changes include proper isolation (skipRegistry, temp HOME override) to prevent cross-test registry pollution.

Confidence Score: 4/5

• Safe to merge with minor verification needed for workflow behavior
• Large refactor is well-structured with comprehensive test coverage (385 tests passing). All changes are backward-compatible with proper exports. The publish workflow consolidation is a significant change that should be verified via test trigger, but the logic is sound. Registry TTL and worktree hooks add valuable safety features. Native engine path normalization fixes real cross-platform bugs.
• Verify .github/workflows/publish.yml triggers correctly for both dev and stable flows via workflow_dispatch test runs before relying on automated publishes

Important Files Changed

Filename	Overview
.github/workflows/publish.yml	Consolidated dev+stable publish into single workflow with auto-versioning; dev releases publish on every main commit with -dev.SHA suffix and dev npm tag
src/parser.js	Refactored from 1917 line monolith to thin wrapper that re-exports extractors from src/extractors/ — all language-specific logic moved to dedicated files
src/extractors/javascript.js	Extracted JavaScript/TypeScript parser logic from parser.js with no functional changes, just module reorganization
src/cli.js	Added stats command for graph health overview, auto-prune on registry list, changed default embed model to jina-code, dynamic version from package.json
src/queries.js	Added stats and statsData functions for comprehensive graph health metrics (nodes, edges, languages, cycles, hotspots, embeddings); fixed git repo detection in diffImpact
src/registry.js	Added TTL-based pruning (30-day default), lastAccessedAt tracking, CODEGRAPH_REGISTRY_PATH env var support, preserves addedAt on re-registration
src/builder.js	Skip auto-registration for temp directories, added skipRegistry option for test isolation
src/resolve.js	Normalize import paths from native engine to fix cross-platform path inconsistencies, guard against missing aliases.paths
crates/codegraph-core/src/import_resolution.rs	Improved path normalization to clean . and .. segments via PathBuf components, ensuring cross-platform consistency
.claude/hooks/guard-git.sh	New PreToolUse hook that blocks dangerous git commands (add -A, reset, clean, stash) and validates commits against session edit log for parallel session safety
.claude/hooks/track-edits.sh	New PostToolUse hook that logs all Edit/Write operations to session-edits.log for commit validation by guard-git.sh; uses node instead of jq for Windows compatibility
tests/unit/registry.test.js	Comprehensive test coverage for TTL pruning, lastAccessedAt tracking, env var override, and addedAt preservation
tests/search/embedding-regression.test.js	New real-model integration test that validates semantic search quality with 5 regression queries, skips gracefully when transformers not installed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Main Push to main] --> B{Event Type?}
    B -->|push| C[Dev Publish Flow]
    B -->|release| D[Stable Release Flow]
    
    C --> E[Compute Dev Version<br/>MAJOR.MINOR.PATCH+1-dev.SHA]
    D --> F[Extract Version from Tag]
    
    E --> G[Run Tests]
    F --> G
    
    G --> H[Build Native Binaries<br/>Linux/macOS/Windows]
    H --> I[Publish to npm]
    
    I --> J{Is Stable?}
    J -->|Yes| K[Create Version Bump PR<br/>+ Push Git Tag]
    J -->|No| L[Skip PR/Tag<br/>Show Dev Install Instructions]
    
    subgraph "Parser Refactor"
        M[parser.js<br/>1917 lines] -.refactor.-> N[parser.js<br/>thin wrapper]
        N --> O[extractors/javascript.js]
        N --> P[extractors/python.js]
        N --> Q[extractors/go.js]
        N --> R[extractors/rust.js]
        N --> S[extractors/java.js]
        N --> T[extractors/...]
        O & P & Q & R & S & T --> U[extractors/helpers.js]
    end
    
    subgraph "Worktree Hooks"
        V[Edit/Write] --> W[track-edits.sh<br/>Log to session-edits.log]
        W --> X[rebuild-graph.sh<br/>Incremental rebuild]
        Y[git commit] --> Z[guard-git.sh<br/>Validate staged files]
        Z -.check.-> W
    end
    
    subgraph "Registry TTL"
        AA[registerRepo] --> AB[Set lastAccessedAt]
        AC[resolveRepoDbPath] --> AB
        AD[pruneRegistry] --> AE{Check each entry}
        AE -->|Missing dir| AF[Remove: missing]
        AE -->|Idle > 30 days| AG[Remove: expired]
        AE -->|Active| AH[Keep]
    end

Loading

_{Last reviewed commit: ac0b198}

[greptile-apps]

_{45 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}