Every task artifact: one reusable, composable, cacheable step of the workflow engine - set a tool up, run a command, clone a repository, download a file, launch an agent - declared in _desc.yaml and run with cx task run <name>; a task lists other tasks under uses, which is how automation composes.
The commands:
cx task listcx task run <name> [--param=value] [-- <passthrough>]cx task run <name> --infocx task run <name> -jcx repo get cTuningLabs@cmeta-aops) and every row below is a command on your machine.
The category itself - its commands and API - is here.
| Artifact | Comes from | What it is for | Tags |
|---|---|---|---|
| add-border-to-images | (cTuningLabs@cmeta-aops) | Add a border to images with a small Python script: sets up Python and Pillow, then runs src/app.py. | |
| add-cmeta-artifacts-to-obsidian | (cTuningLabs@cmeta-aops) | Export the cMeta artifacts of a category into an Obsidian vault as linked notes. | |
| analyze-disk-usage-linux | (cTuningLabs@cmeta-aops) | Show what fills the disk: ncdu on Linux and macOS (installed as a system package), WinDirStat on Windows. | setup x11 |
| attach-tmux | (cTuningLabs@cmeta-aops) | Attach to the running tmux session, setting tmux up first (not on Windows). | start tailscale service |
| classify-image-pytorch-cuda | (cTuningLabs@cmeta-aops) | Classify an image with PyTorch on CUDA (early skeleton; the image-classification-pytorch program is the working path). | |
| clone-git | (cTuningLabs@cmeta-aops) | Clone a git repository into a directory, setting git up first: cx task run clone-git <url> [<directory>]. | clone git |
| clone-git-to-cache | (cTuningLabs@cmeta-aops) | Clone a git repository into the cMeta cache once and reuse it by name across tasks; the MLPerf sources and results use it. | clone git |
| cmd | (cTuningLabs@cmeta-aops) | Run a shell command with the environment aggregated by the previous steps; the building block most tasks end with. | run cmd |
| compile-and-run-program | (cTuningLabs@cmeta-aops) | Select a program, set up its target and compiler, compile and run it: the engine behind cx program run. | compile-and-run program |
| compile-program | (cTuningLabs@cmeta-aops) | Compile a program without running it (compile-and-run-program in compile-only mode). | compile program |
| compiler | (cTuningLabs@cmeta-aops) | Select and set up a compiler for a language and compute target (C, C++, CUDA, Rust, ...) and record it for the following steps. | |
| dataset | (cTuningLabs@cmeta-aops) | Resolve a dataset artifact by name or tags and expose its files to the program being run. | |
| delete-files | (cTuningLabs@cmeta-aops) | Delete the given files or directories, portably. | |
| detect-python-env | (cTuningLabs@cmeta-aops) | Detect the active Python environment (venv, conda or system) and record it for later steps. | detect python env |
| download-file | (cTuningLabs@cmeta-aops) | Download a file (or the first reachable of several mirrors) into a directory, verify its checksum and unpack archives on request. | download file df |
| enable-long-paths-win | (cTuningLabs@cmeta-aops) | Enable long file paths in the Windows registry so deep artifact trees do not break tools (Windows only). | enable long paths win |
| expand-tmp-to-full-disk | (cTuningLabs@cmeta-aops) | /tmp moved from a RAM-backed tmpfs with a per-user quota (Ubuntu 26.04, Fedora, Arch) to the disk of the root filesystem, lasting and at once, so that large downloads and builds that unpack in /tmp no longer stop with "No space left" or "Disk quota exceeded"; reports on macOS, stops on Windows | task setup tmp disk linux |
| explore-tool-versions | (cTuningLabs@cmeta-aops) | try a tool at several versions - its setup and, when asked, a program with it - in fresh cache entries, record a verdict per version, and propose the next default_version or a limit of latest from the verdicts | task versions explore reproducibility |
| finish-run | (cTuningLabs@cmeta-aops) | Close a program run: collect outputs and timings into the run record. | |
| force-fail | (cTuningLabs@cmeta-aops) | Stop a pipeline with a message when a condition holds, for example an unsupported operating system. | force fail |
| generate-temp-file | (cTuningLabs@cmeta-aops) | Create a temporary file and hand its path to the following steps. | generate temp file |
| get-directory-size | (cTuningLabs@cmeta-aops) | Print the size of a directory (used to watch how source trees grow). | get git config ggc |
| get-git-config | (cTuningLabs@cmeta-aops) | Print the git configuration of a repository or of the user. | get git config ggc |
| get-hf-dataset | (cTuningLabs@cmeta-aops) | Download a dataset from the Hugging Face Hub into the cache with the hf CLI, after hf auth login when asked. | |
| get-hf-model | (cTuningLabs@cmeta-aops) | Download a model from the Hugging Face Hub into the cache with the hf CLI, after hf auth login when asked. | |
| get-mlperf-endpoints-results | (cTuningLabs@cmeta-aops) | Clone the MLPerf Inference Endpoints results repository of a given round (endpoints_results_v<round>) into the cache. | get results mlperf endpoints inference github mlcommons |
| get-mlperf-endpoints-src | (cTuningLabs@cmeta-aops) | Clone the MLPerf Inference Endpoints benchmark source (mlcommons/endpoints) into the cache. | get src mlperf endpoints inference github mlcommons |
| get-mlperf-inference-results | (cTuningLabs@cmeta-aops) | Clone the MLPerf Inference results repository of a given round into the cache. | |
| get-mlperf-inference-src | (cTuningLabs@cmeta-aops) | Clone the MLPerf Inference source repository (mlcommons/inference) into the cache. | |
| get-mlperf-training-results | (cTuningLabs@cmeta-aops) | Clone the MLPerf Training results repository of a given round into the cache. | |
| host | (cTuningLabs@cmeta-aops) | Detect the host: operating system, architecture, shell conventions and the variables every other task relies on. | use host |
| import-ai | (cTuningLabs@cmeta-aops) | Import memories and skills (Claude Code memory files, repositories' .claude/skills) into the !AI folder of a cMeta artifact, for run-ai. | ai memory skills import run-ai |
| init | (cTuningLabs@cmeta-aops) | Initialise the run: default settings and the flags shared by the whole pipeline. | init host |
| install-sys-tool | (cTuningLabs@cmeta-aops) | Install a system package by name (apt, dnf, apk, brew, ...) with the right command for this operating system. | install sys-tool |
| keep-adb-connection-alive | (cTuningLabs@cmeta-aops) | Reconnect adb to a device over Wi-Fi at a given address every few seconds so long sessions do not drop. | |
| list-git-branches | (cTuningLabs@cmeta-aops) | List all branches of a git repository. | |
| list-git-tags | (cTuningLabs@cmeta-aops) | List the tags of a remote git repository from its URL. | |
| list-wsl | (cTuningLabs@cmeta-aops) | List the installed Windows Subsystem for Linux distributions and their state (Windows only). | pull git pr pgpr |
| md5sum | (cTuningLabs@cmeta-aops) | Compute the MD5 checksum of a file, portably. | |
| model | (cTuningLabs@cmeta-aops) | Resolve a model artifact by name or tags and expose its files to the program being run. | |
| prevent-mac-to-sleep | (cTuningLabs@cmeta-aops) | Disable sleep on macOS (sudo pmset) so long runs survive an idle laptop. | prevent mac sleep |
| print-current-git-branch | (cTuningLabs@cmeta-aops) | Print the current branch of a git repository. | print current git branch pcgb |
| print-exe-deps | (cTuningLabs@cmeta-aops) | Print the shared-library dependencies of an executable: dumpbin on Windows, ldd on Linux, dyld_info on macOS. | |
| print-text | (cTuningLabs@cmeta-aops) | Print a text; a building block for messages inside pipelines. | force fail |
| pull-git-pr | (cTuningLabs@cmeta-aops) | Fetch a GitHub pull request by number into a local branch and check it out. | pull git pr pgpr |
| push-git-branch | (cTuningLabs@cmeta-aops) | Push a branch to origin. | push git branch |
| rclone-to-ssh | (cTuningLabs@cmeta-aops) | Sync, copy or bisync a local directory to a plain SSH host with rclone through an on-the-fly sftp remote: no rclone.conf entry, no stored credentials. | run rclone sync copy bisync backup ssh sftp |
| reboot | (cTuningLabs@cmeta-aops) | Confirm and reboot Windows, Linux or macOS; Linux systemd hosts can override session inhibitors. | reboot system power |
| remove-conda-from-cmd-autorun-win | (cTuningLabs@cmeta-aops) | Remove the conda hook from the cmd.exe AutoRun so new consoles start clean (Windows only). | enable long paths win |
| run-agy | (cTuningLabs@cmeta-aops) | Run Antigravity CLI (agy) on a prompt from --prompt/--prompt_file, headless (agy -p) or interactive; records the output and token stats. | run agy antigravity antigravity-cli google ai cli prompt |
| run-ai | (cTuningLabs@cmeta-aops) | Run a coding agent (Claude Code, Codex, OpenCode, OpenClaw, Antigravity CLI, Gemini CLI) on a cMeta artifact, with its memory, skills, logs and conversations kept in the artifact's !AI folder; continues the last conversation with any harness and model. | run ai agent memory skills conversations claude codex |
| run-az | (cTuningLabs@cmeta-aops) | Set up the Azure CLI and run it: everything after -- goes to az, with the terminal attached so that az login works. | run az azure azure-cli microsoft cloud cli |
| run-claude | (cTuningLabs@cmeta-aops) | Run Claude Code on a prompt from --prompt/--prompt_file, headless (claude -p) or interactive; records the output and token stats; adds cMeta repositories to its context with --add-dir. | run claude ai cli prompt non-interactive headless interactive |
| run-codex | (cTuningLabs@cmeta-aops) | Run OpenAI Codex on a prompt from --prompt/--prompt_file, headless (codex exec) or interactive; records the output and token stats. | run codex openai ai cli prompt non-interactive headless |
| run-gemini | (cTuningLabs@cmeta-aops) | Run Gemini CLI on a prompt from --prompt/--prompt_file, headless (gemini -p) or interactive; records the output and token stats. | run gemini gemini-cli google ai cli prompt non-interactive |
| run-hermes | (cTuningLabs@cmeta-aops) | Run Hermes Agent (hermes) on a prompt from --prompt/--prompt_file, headless (hermes chat -Q -q) or interactive; records the output, the session id and the token stats. | run hermes hermes-agent nous ai cli prompt non-interactive |
| run-openclaw | (cTuningLabs@cmeta-aops) | Run the OpenClaw agent on a prompt from --prompt/--prompt_file, headless (openclaw agent --local) or in its terminal UI; records the output and token stats. | run openclaw ai agent cli prompt non-interactive headless |
| run-opencode | (cTuningLabs@cmeta-aops) | Run OpenCode.AI on a prompt from --prompt/--prompt_file, headless (opencode run) or interactive; records the output and token stats. | run opencode opencode.ai ai cli prompt non-interactive headless |
| run-openssh | (cTuningLabs@cmeta-aops) | Start, stop or show an OpenSSH server run as the user (no service, no administrator) on its own port, with key-only logins, on the Tailscale address by default. | run ssh sshd openssh server |
| run-program | (cTuningLabs@cmeta-aops) | Run an already compiled program (compile-and-run-program in run-only mode). | run program |
| runner | (cTuningLabs@cmeta-aops) | Bootstrap a pipeline: host detection plus the shared environment; the first step of almost every task. | init runner |
| scan-git-secrets | (cTuningLabs@cmeta-aops) | Scan a git history (or a folder) for secrets with gitleaks; report file, line, rule and commit - never the values; exit 1 on findings. | scan secrets security gitleaks git hygiene |
| select-program | (cTuningLabs@cmeta-aops) | Find the program artifact to run by name and compute target and load its recipe. | select program |
| select-src-by-tool-version | (cTuningLabs@cmeta-aops) | Select the source folder of a program by the version of a tool set up earlier in the run, so that one program keeps sources for several generations of a language or library. | select src program version |
| set-git-to-main-branch | (cTuningLabs@cmeta-aops) | Set the upstream of a branch to origin/main (or another branch of origin). | set git to main branch |
| set-local-vars | (cTuningLabs@cmeta-aops) | Set local variables for the following steps of a pipeline. | |
| setup | (cTuningLabs@cmeta-aops) | Detect, install or build a tool by name with version matching and caching: the task behind cx tool setup. | setup tool |
| setup-amd-gpu | (cTuningLabs@cmeta-aops) | a Linux machine made ready to compute on its AMD GPU and NPU - the user's access to the device nodes (render and video groups, at once through an ACL), the memory-lock limit the NPU runtime needs - with what was found and what was changed | task setup amd gpu npu rocm linux |
| setup-cmeta-host | (cTuningLabs@cmeta-aops) | Select and prepare optional host services for cMeta fleet access on Windows, Linux and macOS. | setup host fleet ssh tailscale |
| setup-compile | (cTuningLabs@cmeta-aops) | Set up everything needed to compile for a language and compute target: the target, then its compiler. | |
| setup-dynamic-libs | (cTuningLabs@cmeta-aops) | Set up the dynamic libraries a program needs and add their paths to the environment. | |
| setup-git | (cTuningLabs@cmeta-aops) | Detect or install git. | setup git sg |
| setup-git-branch | (cTuningLabs@cmeta-aops) | Create and check out a new git branch. | setup git branch sgb |
| setup-min-win-tools | (cTuningLabs@cmeta-aops) | Download the minimal set of Unix tools for Windows (tar, gzip, md5sum, ...) from Zenodo so archives and checksums work there. | download min win tools |
| setup-run | (cTuningLabs@cmeta-aops) | Set up the run environment of a program on a compute target and execute it, collecting outputs. | |
| setup-x11 | (cTuningLabs@cmeta-aops) | Install the X11 client packages (xauth, xeyes, x11-apps or xterm) so GUI tools work over SSH on Linux. | setup x11 |
| start-tailscale-service | (cTuningLabs@cmeta-aops) | Start the Tailscale service (macOS, through brew services). | start tailscale service |
| start-tmux | (cTuningLabs@cmeta-aops) | Start a tmux session, setting tmux up first (not on Windows). | start tailscale service |
| stop-wsl | (cTuningLabs@cmeta-aops) | Stop the Windows Subsystem for Linux service (Windows only). | pull git pr pgpr |
| sync-system-clock | (cTuningLabs@cmeta-aops) | the system clock checked against a time server and put right - the time service restarted and resynced - with the OS's own commands | task clock time ntp system |
| system | (cTuningLabs@cmeta-aops) | Expose the detected system information to a pipeline (a thin wrapper over runner). | use system |
| target | (cTuningLabs@cmeta-aops) | Select the compute target (cpu, cuda, rocm, metal, xpu, android-cpu, ...) and dispatch to its target--* task. | |
| target--android-cpu | (cTuningLabs@cmeta-aops) | Compute target: an Android device over adb, selected by serial; the NDK toolchain and sysroot follow. | |
| target--android-gpu | (cTuningLabs@cmeta-aops) | Compute target: the GPU of an Android device over adb (the device as android-cpu selects it); records its Vulkan devices, GLES driver and OpenCL library; NDK programs run there. | target android gpu |
| target--android-npu | (cTuningLabs@cmeta-aops) | Compute target: the NPU of an Android device over adb (the device as android-cpu selects it) through NNAPI; records the NNAPI accelerators and the vendor NPU stack (Samsung ENN, MediaTek APU, Google EdgeTPU, Qualcomm Hexagon). | target android npu |
| target--cpu | (cTuningLabs@cmeta-aops) | Compute target: the host CPU. | |
| target--cuda | (cTuningLabs@cmeta-aops) | Compute target: NVIDIA GPU through the detected CUDA toolkit. | |
| target--metal | (cTuningLabs@cmeta-aops) | Compute target: Apple GPU through Metal (macOS only; records the GPU from system_profiler). | |
| target--npu-amd | (cTuningLabs@cmeta-aops) | Compute target: the AMD Ryzen AI NPU (XDNA) through FastFlowLM; records the NPU, its PCI ID, generation, firmware and device node, without any SDK. | target npu amd |
| target--npu-intel | (cTuningLabs@cmeta-aops) | Compute target: the Intel NPU (Meteor Lake and newer) through OpenVINO; records the NPU, its PCI ID, generation and driver, without any SDK. | target npu intel |
| target--opencl | (cTuningLabs@cmeta-aops) | Compute target: a GPU through OpenCL (any vendor); records the platforms and devices; on Linux, an Intel GPU gets Intel's runtime first. | target opencl gpu |
| target--openvino | (cTuningLabs@cmeta-aops) | Compute target: the OpenVINO stack for Intel CPUs, GPUs and NPUs; combine it with a device target (npu-intel, xpu, cpu), or use it alone for every device OpenVINO finds. | target stack openvino |
| target--opu-lumai | (cTuningLabs@cmeta-aops) | Compute target: LumAI OPU (placeholder, not yet supported). | |
| target--rocm | (cTuningLabs@cmeta-aops) | Compute target: AMD GPU through the detected ROCm platform. | |
| target--tpu | (cTuningLabs@cmeta-aops) | Compute target: Google TPU (placeholder, not yet supported). | |
| target--vulkan | (cTuningLabs@cmeta-aops) | Compute target: any GPU through Vulkan (NVIDIA, AMD, Intel, Apple via MoltenVK, Mesa's llvmpipe); records the Vulkan devices; the Vulkan SDK follows for builds. | |
| target--xpu | (cTuningLabs@cmeta-aops) | Compute target: Intel GPU (XPU) through oneAPI; records the graphics adapter. | |
| target-sdk | (cTuningLabs@cmeta-aops) | Set up the SDK of the selected compute target (CUDA, ROCm, Metal, Android NDK) before compiling. | |
| test-adb | (cTuningLabs@cmeta-aops) | Smoke test: set adb up and list the connected Android devices. | |
| test-android-clang-compile-and-run | (cTuningLabs@cmeta-aops) | Smoke test: cross-compile a C program with the Android NDK clang and run it on the connected device. | |
| test-any-cpp | (cTuningLabs@cmeta-aops) | Smoke test: compile and run a C++ program with whichever C++ compiler the compiler task selects. | |
| test-build-pytorch-from-git | (cTuningLabs@cmeta-aops) | Build PyTorch from its git source with the detected compilers, CMake and Ninja (setup.py route) for the selected target. | |
| test-build-pytorch-from-git-cmake | (cTuningLabs@cmeta-aops) | Build PyTorch from its git source through a direct CMake configuration for the selected target. | |
| test-build-pytorchvision-from-git | (cTuningLabs@cmeta-aops) | Build torchvision from git against the PyTorch built from source. | |
| test-clang-cpp | (cTuningLabs@cmeta-aops) | Smoke test: detect or install the Clang C++ compiler. | |
| test-claude | (cTuningLabs@cmeta-aops) | Smoke test: set Claude Code up and ask it one question in print mode (-p). | |
| test-coreml-clang | (cTuningLabs@cmeta-aops) | Smoke test: compile and run an Objective-C++ program against the Core ML framework on macOS. | |
| test-coreml-python | (cTuningLabs@cmeta-aops) | Smoke test: run coremltools from Python (Python below 3.13) on macOS. | |
| test-gcc-cpp | (cTuningLabs@cmeta-aops) | Smoke test: detect or install the GNU C++ compiler (not on Windows). | |
| test-go-build-and-run | (cTuningLabs@cmeta-aops) | Smoke test: build a Go hello-world binary and run it. | |
| test-go-run | (cTuningLabs@cmeta-aops) | Smoke test: go run a hello-world program. | |
| test-icpx | (cTuningLabs@cmeta-aops) | Smoke test: compile and run a C++ program with the Intel oneAPI icpx compiler (Linux and Windows). | |
| test-icpx-sycl-gpu | (cTuningLabs@cmeta-aops) | Smoke test: compile and run a SYCL program with icpx -fsycl on an Intel GPU. | |
| test-java-compile-and-run | (cTuningLabs@cmeta-aops) | Smoke test: compile a Java hello-world with javac and run it with java. | |
| test-mojo | (cTuningLabs@cmeta-aops) | Smoke test: run a hello-world with the Mojo compiler. | |
| test-mojo-life | (cTuningLabs@cmeta-aops) | Run the Mojo Game of Life example with its pygame-ce front end and benchmark. | |
| test-msvc | (cTuningLabs@cmeta-aops) | Smoke test: compile and run a C++ program with Microsoft cl.exe (Windows only). | |
| test-node-js | (cTuningLabs@cmeta-aops) | Smoke test: run a hello-world with Node.js. | |
| test-node-js-modular-skill-TBD | (cTuningLabs@cmeta-aops) | Experiment: install the Modular skills with npx and run a Node.js test (work in progress). | |
| test-nvcc | (cTuningLabs@cmeta-aops) | Smoke test: detect the CUDA compiler nvcc and add it to the environment. | |
| test-python | (cTuningLabs@cmeta-aops) | Smoke test: detect or set up Python (with a virtual environment by default). | |
| test-python-gradio | (cTuningLabs@cmeta-aops) | Smoke test: install Gradio and start its demo app on a local port. | |
| test-python-numpy | (cTuningLabs@cmeta-aops) | Smoke test: install NumPy and run a small script. | |
| test-python-pygame | (cTuningLabs@cmeta-aops) | Smoke test: install pygame and open a small window. | |
| test-python-streamlit | (cTuningLabs@cmeta-aops) | Smoke test: install Streamlit and run streamlit hello. | |
| test-pytorch | (cTuningLabs@cmeta-aops) | Smoke test: install PyTorch for the selected compute target and run a small script. | |
| test-pytorch-cpu | (cTuningLabs@cmeta-aops) | Smoke test: install PyTorch for the CPU target and run a small script. | |
| test-pytorch-cpu-torchvision | (cTuningLabs@cmeta-aops) | Smoke test: install PyTorch, torchvision and torchaudio for the CPU target and run a small script. | |
| test-pytorch-cuda | (cTuningLabs@cmeta-aops) | Smoke test: install PyTorch for CUDA and run a small script on the GPU. | |
| test-pytorch-metal | (cTuningLabs@cmeta-aops) | Smoke test: install PyTorch for Apple Metal (MPS) and run a small script. | |
| test-pytorch-rocm | (cTuningLabs@cmeta-aops) | Smoke test: install PyTorch for AMD ROCm and run a small script on the GPU. | |
| test-pytorch-xpu | (cTuningLabs@cmeta-aops) | Smoke test: install PyTorch for Intel XPU (with the oneAPI icx compiler) and run a small script. | |
| test-rustc-build-and-run | (cTuningLabs@cmeta-aops) | Smoke test: compile a Rust hello-world with rustc and run it. | |
| test-session | (cTuningLabs@cmeta-aops) | Test sessions: a dated sandbox for a test, build or benchmark and a record that stays (folders of tmp:: and log::cmeta-aops-test-sessions in the local repository): what was done, by which agent, model and effort, the results and the costs. | test session sandbox log |
| test-start-vllm | (cTuningLabs@cmeta-aops) | Smoke test: install vLLM for CUDA and start it from a small script. | |
| test-swiftlang | (cTuningLabs@cmeta-aops) | Smoke test: run a Swift hello-world with the Swift toolchain. | |
| test-swiftlang-build-and-run | (cTuningLabs@cmeta-aops) | Smoke test: compile a Swift hello-world with swiftc and run it. | |
| ungzip | (cTuningLabs@cmeta-aops) | Decompress a .gz file into a directory, setting gzip up first. | ungzip file |
| untar | (cTuningLabs@cmeta-aops) | Extract a tar archive into a directory, setting tar up first. | untar file |
| unzip | (cTuningLabs@cmeta-aops) | Extract a zip archive into a directory, setting unzip up first. | unzip file |
| update-git-submodules | (cTuningLabs@cmeta-aops) | Sync and update the submodules of a git repository recursively. | |
| upgrade-os | (cTuningLabs@cmeta-aops) | the OS packages (and firmware when asked) brought up to date with the OS's own commands, and whether a reboot is due | task upgrade os system firmware |
| venv | (cTuningLabs@cmeta-aops) | Create or reuse a Python virtual environment with uv. | setup uv venv |
| zip | (cTuningLabs@cmeta-aops) | Create a zip archive of a directory, portably. | zip file |
Nothing matches that filter.
Read live from the cMeta index of this server - the same data cx category list shows you locally.