Convert static imports of @huggingface/transformers (which triggers
onnxruntime-node native binary loading) to dynamic import() calls.
This prevents crashes on Node versions whose ABI isn't supported by
the prebuilt onnxruntime binaries (e.g. Node v24).
Affected entry points:
- cli/analyze.ts: embedding pipeline only loaded when --embeddings is passed
- mcp/local/local-backend.ts: embedder only loaded on first semantic search
- server/api.ts: embedder only loaded when search endpoint needs embeddings
Fixes#89
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The MCP server called process.exit(1) at startup when no repositories
were found in the registry. This prevented users from configuring the
MCP integration before running `gitnexus analyze`.
The server now starts gracefully with 0 repos and discovers newly
indexed repos lazily via refreshRepos() on each tool call.
Closes#91
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Keep main's barLog implementation, preserve both currentDbPath and
ftsLoaded reset in closeKuzu, take PR's new resolveRepo pattern
for /api/query. Path traversal guard confirmed intact.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Skill descriptions were too tool-centric ("using knowledge graph", "blast
radius") which prevented Claude Code from matching them to user intent.
Rewritten to user-intent-driven format with "Use when..." phrasing and
example trigger phrases so Claude can semantically match user requests.
Updated across all 3 sources: gitnexus/skills/, gitnexus-claude-plugin/skills/,
.claude/skills/, and the ai-context.ts fallback generator.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Address code review feedback on the server-mode PR:
Critical fixes:
- Restore CORS whitelist (localhost + gitnexus.vercel.app only)
- Bind to 127.0.0.1 by default; add --host CLI flag for opt-in remote access
- Restore path traversal guard on /api/file (resolve + startsWith check)
- Restore try/catch on all route handlers + global error middleware
- Restore SIGINT/SIGTERM graceful shutdown handlers
Bug fixes:
- Add mutex to core initKuzu to prevent race conditions on concurrent
DB switches (two requests for different repos no longer corrupt state)
- Track ftsLoaded flag and reset on DB switch / close so FTS extension
is reloaded for each new database connection
- Restore input validation on /api/query (cypher required) and
/api/search (query required)
Improvements:
- Add TTL-based cleanup for orphaned MCP sessions (30min idle eviction)
to prevent memory leaks from network drops that skip onclose
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Raise MAX_FILE_SIZE from 512KB to 2MB to capture more real source files
- Replace verbose per-warning output with single summary line
- Soften skip message wording ("likely generated/vendored")
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Skill folder names determine invocation paths in Claude Code plugins
(e.g. plugin:gitnexus:gitnexus-cli). Generic names like "cli" or
"debugging" could collide with other plugins, so prefix them all with
gitnexus- for clarity.
Updated across plugin dirs, main package source files, ai-context.ts
generator, setup.ts installer, and all CLAUDE.md/AGENTS.md routing tables.
- Impact tool now returns risk score, affected processes/modules, and summary
- Cypher tool formats results as markdown tables for LLM readability
- Context tool includes module (functional area) field
- Semantic search skips model init when embeddings are disabled
- Setup: wrap npx in cmd /c on Windows for .cmd script compatibility
- Embedder: silence stderr during ONNX model load to protect MCP stdio
- API: use executeCypher directly to avoid double formatting
- Add community integrations section to READMEs
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Bundle MCP server config (.mcp.json), fix hook to use spawnSync with npx
fallback and read stderr (KuzuDB stdout workaround), wire plugin.json with
hooks/mcpServers paths, add guide skill with tools/resources/schema reference,
add AMP-compatible mcp.json to all skill dirs, and slim CLAUDE.md generator
by moving reference content into the guide skill.
Replace the old local-only backend mode with a new server connection
flow that lets the web UI connect to any running GitNexus server,
download the pre-built knowledge graph, and explore it without WASM.
- Add server-connection service with streaming download and progress
- Replace DropZone backend tab with server connect UI (URL input, progress bar, cancel)
- Add repo switcher dropdown in Header when multiple repos are indexed
- Mount MCP server over StreamableHTTP at /api/mcp for remote AI tool access
- Refactor api.ts to query KuzuDB directly instead of routing through LocalBackend
- Fix KuzuDB adapter to support switching between databases (close old before opening new)
- Extract createMCPServer() from startMCPServer() for transport-agnostic reuse
- Support ?server= query param for bookmarkable auto-connect
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Previous usage was vulnerable to "poisoned" tags
which could enduce commands to be run when a
`detectChanges` command was hit. This was primarily
fixed in `local-backend.ts` however I changes the
`execSync` usages where any injection was potentially
able to be performed (e.g. staleness).
Skipped touching wiki and generator as those use static
input, though these should potentially be changed over
in the future.
SingleBar from cli-progress doesn't have a .log() method (that's
MultiBar-only), causing a crash when any code calls console.log
during the pipeline.
Fixes#53
Thanks to @TranDatk for reporting in #56.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Embeddings (ONNX Runtime) caused segfaults on macOS (#38) and CUDA
install failures on Linux (#40). Since embeddings only power semantic
vector search in query() and BM25 + graph traversal already provide
strong retrieval, disable embeddings by default with minimal quality
impact.
Changes:
- Replace --skip-embeddings with --embeddings opt-in flag
- Remove disposeEmbedder() call (ONNX native cleanup segfaults)
- Add process.exit(0) when embeddings are used to bypass ONNX atexit hooks
- Suppress ONNX session warnings via logSeverityLevel
- Fix progress bar duplication by routing console output through bar.log()
- Add elapsed time indicator for long-running phases
Fixes#38Fixes#40
listRepos() now re-reads ~/.gitnexus/registry.json on every call.
resolveRepo() retries with a registry refresh on cache miss.
Extracted refreshRepos() from init() for reuse. All callers updated
with await for the new async signatures.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Hook config goes in ~/.claude/settings.json (not hooks.json)
- Matcher uses string format ("Grep|Glob|Bash") per new Claude Code schema
- Rename gitnexus-hook.js → gitnexus-hook.cjs for CommonJS compatibility
- Fix setup.ts: correct hook filename and timeout (8000ms instead of 10ms)
- Bump to v1.1.9 and publish to npm
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
