Clawhub

Use when the user wants to run GPU-accelerated computations, analyze data, process images/audio/video, train ML models, or generate code from natural languag...

Security Scan

The skill is internally consistent with a localized GPU-tool that delegates work to a local 'octoflow' binary; nothing requested is disproportionate to its stated purpose, though there are minor install metadata quirks to verify before use.

Purpose & Capability

The name/description match the declared runtime requirement (an 'octoflow' binary) and the SKILL.md instructions show only operations relevant to GPU compute, data analysis, and running an MCP server. Requiring a local octoflow binary is reasonable for this functionality.

Instruction Scope

Runtime instructions are narrowly scoped: run the octoflow binary, pass permission flags, run as an MCP server for agent integration, and operate on user-permitted files/domains. The SKILL.md adheres to a deny-by-default permission model and does not instruct reading unrelated system files or exfiltrating secrets.

Install Mechanism

The SKILL.md includes an install hint pointing to GitHub Releases (a normal source). However the provided URL ends at the release directory (no specific asset filename) which is an inconsistency and would need resolution by an installer. Overall, using GitHub Releases is expected and low risk compared to arbitrary hosts.

Credentials

The skill declares no required environment variables or credentials. It does note that OctoFlow may save preferences under ~/.octoflow and per-project .octoflow/ which is plausible and proportionate to the tool's function.

Persistence & Privilege

always is false and the skill does not request permanent elevated platform privileges. It can run a local MCP server (expected for agent integration) — this grants a local RPC surface only when the user starts it or allows the agent to. The deny-by-default flags limit automatic access.

Assessment

This skill appears to be what it claims: a front-end to a local 'octoflow' GPU binary. Before installing or running it: (1) ensure you obtain the octoflow binary from the official GitHub release page and verify the exact asset name and checksum (the SKILL.md's install URL is missing the asset filename); (2) be deliberate when granting --allow-read / --allow-write / --allow-net / --allow-exec flags — only allow the minimal paths/domains/commands needed; (3) running octoflow mcp-serve exposes a local RPC endpoint to your agent — only enable that if you trust the binary and your agent config; (4) expect OctoFlow to write small preference files to ~/.octoflow and per-project .octoflow/; and (5) if you need stronger assurance, verify the release signature/checksum on GitHub or build the binary from source.

OctoFlow

GPU-native programming language. Describe tasks in English, run them on any GPU via Vulkan.

When to Use This Skill

Use this skill when the user says:

• "sort a million numbers on GPU" / "benchmark GPU performance"
• "load this CSV and show me statistics" / "analyze my data"
• "cluster my dataset" / "run K-means" / "train a classifier"
• "blur this image" / "resize this BMP" / "encode a GIF"
• "plot height vs weight" / "create a scatter plot"
• "calculate the Sharpe ratio" / "compute correlation"
• "find primes" / "generate random numbers on GPU"
• "run linear regression on my dataset"
• "run a LLM on my GPU" / "inference on my GPU"

Do NOT use this skill when:

• The user wants Python/JavaScript/Rust code (use the appropriate language tool)
• The task doesn't benefit from GPU acceleration or OctoFlow's built-in functions
• The user explicitly asks for a different language

How to Run OctoFlow

Chat mode (natural language to running code)

octoflow chat "sort 1M numbers on GPU"

Run a .flow script

octoflow run program.flow

Run with permissions (sandboxed by default)

# Allow reading data files
octoflow run analysis.flow --allow-read=./data

# Allow network access to specific domain
octoflow chat "fetch weather data" --allow-net=api.weather.com

# Allow writing output files
octoflow run report.flow --allow-read=./data --allow-write=./output

Security Model

OctoFlow uses Deno-style permissions. Everything is denied by default.

Permission	Default	How to enable	Example
File read	DENIED	--allow-read=./data	Read CSV from ./data only
File write	DENIED	--allow-write=./output	Write results to ./output only
Network	DENIED	--allow-net=api.example.com	Fetch from one domain only
Process exec	DENIED	--allow-exec=curl	Allow curl only

Without flags, OctoFlow can only read .flow source files and print to stdout. No file access, no network, no subprocesses unless the user explicitly opts in.

MCP Server

OctoFlow can run as an MCP server for AI agent integration:

octoflow mcp-serve

Add to your OpenClaw, Claude Desktop, or Cursor config:

{"mcpServers": {"octoflow": {"command": "octoflow", "args": ["mcp-serve"]}}}

Available Tools

Tool	Description
octoflow_run	Execute OctoFlow code directly
octoflow_chat	Natural language to GPU code
octoflow_check	Validate .flow syntax
octoflow_gpu_sort	GPU-accelerated sorting
octoflow_gpu_stats	GPU statistical operations
octoflow_image	Image processing (BMP, GIF)
octoflow_csv	CSV data analysis

Common Patterns

Data Analysis

# User: "analyze sales.csv and show trends"
octoflow chat "load sales.csv, compute monthly averages, and plot the trend" --allow-read=.

GPU Compute

# User: "sort a large dataset on GPU"
octoflow chat "generate 1M random numbers on GPU and sort them"

Machine Learning

# User: "cluster my customers"
octoflow chat "load customers.csv, run K-means with 5 clusters, print cluster sizes" --allow-read=.

Image Processing

# User: "blur this photo"
octoflow chat "load photo.bmp, apply gaussian blur, save as blurred.bmp" --allow-read=. --allow-write=.

Statistics

# User: "what's the correlation between these columns?"
octoflow chat "load data.csv, compute Pearson correlation between col1 and col2" --allow-read=.

Key Capabilities

Feature	Detail
Builtins	210+ built-in functions
Stdlib	445 modules across 28 domains
GPU kernels	150 Vulkan compute shaders
GPU VM	Loom Engine — 5 SSBOs, indirect dispatch, layer streaming
GPU support	Any Vulkan GPU (NVIDIA, AMD, Intel)
Binary size	4.5 MB, zero dependencies
Chat mode	English to code with auto-fix loop (max 3 retries)
Errors	69 structured error codes with auto-fix suggestions
MCP Server	7 structured tools via JSON-RPC 2.0
Platforms	Windows, Linux, macOS (Apple Silicon)

Data Storage

OctoFlow optionally saves your preferences to ~/.octoflow/ (user-level) and .octoflow/ (per-project).

Contents: which stdlib modules you use frequently and corrections from previous sessions.

• No telemetry. No data is sent anywhere.
• No network calls unless you explicitly use --allow-net.
• All data stays local on your machine.
• Disable entirely with --no-memory flag — nothing is saved.
• Project config via OCTOFLOW.md in your project root (like .eslintrc or pyproject.toml).

Install

Download (recommended)

Platform	File	SHA-256
Windows x64	octoflow-v1.5.8-x86_64-windows.zip	2b26049565a2bfd2b1c4a1c103f2a64cd864dd14da619bd7be750ad3c6b356f2
Linux x64	octoflow-v1.5.8-x86_64-linux.tar.gz	d7306fc1f5a9a733a66ae3a4d5f3b145670efa7a079302935d867b4b75551845
macOS (Apple Silicon)	octoflow-v1.5.8-aarch64-macos.tar.gz	33808c330dc5f08eb0008b52ecfb5f0ea532fb71b1c6996075c09b33dc5d8fd2

Verify: sha256sum octoflow-v1.5.8-* (full checksums in SHA256SUMS.txt).

Unzip/extract, add to PATH. No installer needed.

Links

• GitHub
• Documentation
• Releases