go-ai/TODO.md

TODO — go-ai

Virgil dispatches tasks. Mark [x] when done, note commit hash.

---

Phase 1: Post-Split Cleanup

• Remove test-mlx.go — Deleted standalone test script from module root.
• Verify go build ./... passes — Clean build, no stale import paths.
• Verify go vet ./... passes — No vet warnings.
• Run full test suite — All tests pass. Fixed TestSandboxing_Symlinks_Blocked (renamed; asserts sandbox blocks symlink escape) and TestNewTCPTransport_Warning (added missing security warning to NewTCPTransport).

Phase 2: go-inference Migration

go-ml is migrating to use go-inference shared interfaces. Once that's done, go-ai's ML subsystem should use go-inference too.

• Update tools_ml.go MLSubsystem — mlGenerate/mlScore/mlProbe unchanged (work correctly via go-ml.Service → InferenceAdapter → inference.TextModel). Added flow documentation comments.
• Update ml_backends tool — Rewritten to use inference.List()/Get()/Default() instead of ml.Service.Backends()/Backend()/DefaultBackend().
• Update ml_score and ml_probe — Kept go-ml dependency for scoring/probes (that's where the scoring engine lives). Generation flows through go-inference via InferenceAdapter. Added documentation comments.
• Add go-inference to go.mod — Promoted from indirect to direct require. Replace directive already present.

Phase 3: MCP Transport Testing

• Stdio transport e2e — Test core mcp serve over stdin/stdout with a mock MCP client. Verify tool discovery + file_read round-trip.
• TCP transport e2e — Test MCP_ADDR=:9100 core mcp serve. Connect, list tools, call file_read, verify response.
• Unix transport — Currently untested. Add basic connect + tool call test.
• Webview tools CI guard — tools_webview.go tools require Chrome. Add testing.Short() skip or build tag so CI doesn't fail.

Phase 4: IDE Subsystem Hardening

• Bridge reconnection test — Kill mock Laravel WS server, verify exponential backoff + reconnect in bridge.go.
• Add auth to bridge — bridge.go connects unauthenticated. Add token header on WebSocket upgrade.
• Dashboard tools beyond stubs — tools_dashboard.go returns empty data. Implement real data fetching or document as stub.

Phase 5: Testing Gaps

• Process tools CI tests — tools_process.go needs CI-safe tests (start/stop lightweight processes like echo or sleep).
• RAG tools mock — tools_rag.go needs Qdrant + Ollama mocks for CI. Test rag_query, rag_ingest, rag_collections without live services.
• ML tools mock — tools_ml.go needs mock backend for CI. No real inference in tests.
• Metrics benchmark — Benchmark metrics_record + metrics_query at scale (10K+ JSONL events).

---

Workflow

1. Virgil in core/go writes tasks here after research
2. This repo's session picks up tasks in phase order
3. Mark [x] when done, note commit hash
4. New discoveries → add tasks, flag in FINDINGS.md