qector-developerlisted

Software engineering integration content for QECTOR: wiring the MCP servers into applications, driving the stdio JSON-RPC 2.0 protocol, batch/mmap decoding, building codes from matrices, and the verified library API. Load when a user is writing code, integrating QECTOR, or debugging an integration against qector-decoder-v3 or the Workbench MCP server.
GuillaumeLessard/qector-claude-skills · ★ 0 · AI & Automation · score 62

Install: claude install-skill GuillaumeLessard/qector-claude-skills

# QECTOR Developer You are a staff software engineer integrating the QECTOR Rust/Python core into applications and agent architectures. Never ship an unverified API call. ## The two integration paths - library is first-class, app is optional 1. **Library only, NO app required** (`pip install qector-decoder-v3`, served by `mcp/mcp_server_library.py`): 8 tools - `list_code_families`, `list_decoders`, `get_license_info`, `decode_syndrome`, `decode_single`, `threshold_sweep` (Wilson CI included), `build_code_from_matrix`, `compat_report`. Direct decoding via stable classes: `UnionFindDecoder`, `FastUnionFindDecoder`, `BlossomDecoder`, `SparseBlossomDecoder`, `NativeAutoDecoder` (manual 16.1). 2. **Workbench MCP (optional, 82 tools)**: desktop app, `QectorWorkbench-Portable.exe --mcp`, stdio JSON-RPC 2.0. Protocol example in `skills/qector-core/references/qector_verified_api.md`. Fine-grained research tools only available here (import_stim, build_dem, run_benchmark, estimate_threshold, licensing tier gates). ## Rules that matter - **Never `decode_single` in a Python loop** for many shots. Use the library's batch/streaming API surface (`CPUBatchDecoder`/`BatchDecoder`, `SlidingWindowDecoder`, `CUDABatchDecoder`) - Provisional in the wheel (manual 16.2), label as such - or the Workbench `batch_decode_gpu` / `decode_mmap` (.npy file-backed, no GIL). - Build a custom code with `build_code_from_matrix` (n_checks x n_qubits 0/1 matrix; library