comanda_skill

Overview

This skill lets you generate, visualize, and execute declarative AI pipelines from the command line using comanda. Define multi-step LLM workflows in YAML, chain or parallelize steps, and run them against multiple model providers for end-to-end automation. It supports generating workflows from plain English prompts, viewing workflow charts, editing YAML files, and processing workflows with real inputs.

How this skill works

You author workflows as YAML files where each step declares input, model, action, and output. The CLI can generate those YAML files from natural-language prompts, render an ASCII chart of step relationships and validity, and execute workflows piping data between steps or running them in parallel. comanda integrates with multiple model backends (OpenAI, Anthropic, Google, Ollama, Claude Code, Gemini CLI, Codex) via configured API keys.

When to use it

• You need repeatable LLM pipelines that chain analysis, extraction, and summarization steps.
• You want to quickly prototype a workflow by describing it in natural language.
• You need to visualize or validate complex workflows before running them.
• You must orchestrate multiple models or run steps in parallel for comparison.
• You want to embed model-driven transforms into CI, scripts, or batch processing.

Best practices

• Keep each YAML step focused: single input, a clear instruction, and a specific output variable or file.
• Use comanda chart to validate step dependencies and spot model misconfiguration before execution.
• Store sensitive API keys via comanda configure and avoid hardcoding credentials in YAML files.
• Break large workflows into smaller reusable files and chain them with process commands for clarity.
• Test steps interactively with small inputs before running full datasets or parallel runs.

Example use cases

• Summarize a corpus: extract key points from documents, then generate an executive summary.
• Code review pipeline: run multiple models to analyze code for bugs, suggest fixes, and produce a patch.
• Multi-model comparison: run the same prompt through different providers in parallel to compare outputs.
• Automated document processing: extract text from PDFs, classify sections, and populate structured outputs.
• Meta-workflows: generate a new workflow YAML from a prompt and immediately process the generated workflow.