What happens if a unit fails its acceptance criteria?

If produced artifacts do not meet the criteria, the unit status is set to BLOCKED, and a notification is recorded in STATUS.md for user review.

Why does it only execute one unit at a time?

Executing one unit at a time ensures the process remains auditable and interruptible, preventing the AI from proceeding without valid artifacts or necessary human approval.

What is a 'unit' in this context?

A unit is a single, runnable task defined in a UNITS.csv file, specifying required inputs, expected outputs, and specific acceptance criteria.

Can this skill be used for full automation?

While this skill focuses on single-step execution for safety, it returns specific exit codes (0 for success, 2 for blocked) to support integration into larger automation scripts.

How does the skill handle task dependencies?

It scans the UNITS.csv file and only initiates a task if its status is TODO and all units listed in its 'depends_on' column are marked as DONE.

Unit Pipeline Executor

Name: Unit Pipeline Executor
Author: WILLOSCAR

byWILLOSCAR

•

Productivity & Workflow

Executes structured research or development tasks one at a time with strict dependency management and status tracking.

The Unit Pipeline Executor is a specialized skill designed for controlled, step-by-step processing of research and execution pipelines. By reading from a structured UNITS.csv file, it identifies the next available task, validates its dependencies, executes the required logic, and verifies the output artifacts against specific acceptance criteria. This 'one-at-a-time' approach ensures a high level of auditability, prevents hallucinations through evidence-first methodology, and allows for human-in-the-loop checkpoints, making it ideal for complex workflows where quality and consistency are paramount.

Key Features

01Evidence-first methodology that validates output artifacts before marking tasks as complete

02Real-time status tracking with automated updates to UNITS.csv and STATUS.md

03Strict single-unit execution to ensure auditability and prevent cascading errors

04Automated dependency resolution based on UNITS.csv status mapping

05Built-in support for HUMAN checkpoints and manual approval blocks

0689 GitHub stars

Use Cases

01Implementing auditable AI workflows where each step must be verified before proceeding

02Managing complex multi-step research pipelines that require incremental artifact validation

03Executing long-running development tasks that require periodic human intervention or decision-making

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add willoscar/research-units-pipeline-skills unit-executor

For use in Claude.ai and ChatGPT

Download Skill