GoSkill - An open-source tool for advancing long tasks, with built-in retry loop and state tracking

What is GoSkill

GoSkill is a tool for advancing long tasks and complex tasks. The tool encapsulates tasks via decorators or classes, upgrading "one-off Skill calls" into an execution mode that "continuously advances toward the goal until the success criteria are met or a timeout occurs." GoSkill is suitable for automation scenarios that require clear acceptance criteria, state tracking, and phased completion, such as large-scale refactoring and long-duration analysis.

Main features of GoSkill

• Goal definition: Supports specifying the task goal via the goal parameter, expressing task intent in a structured way.
• Success criteria: Defines multi-dimensional acceptance conditions via a criteria dictionary, quantifying the completion criteria.
• Continuous retries: Built-in execution loop; when the criteria are not met, it automatically waits and keeps trying until success or timeout.
• State tracking: Provides the status attribute to view the number of attempts, runtime, and check results in real time.
• Structured return：run_with_result() Returns a structured result object containing success, attempts, and criteria_report.
• Flexible encapsulation: Supports both the decorator @goskill and the class GoSkill , making it lightweight to embed into existing code.

How to use GoSkill

• Install dependencies: After cloning the repository, run make install-dev to complete environment setup.
• Define the task: Use the decorator or instantiate the GoSkill class, passing in parameters such as goal、criteria、max_hours .
• Write the logic: Implement the specific task logic in a function or lambda and return a result dictionary.
• Execute the run: Call run() to get the raw result, or call run_with_result() to get a structured execution report.
• View state: Use skill.status to view the current goal, number of attempts, elapsed runtime, and criteria status.
• Run the example: Execute python examples/basic_usage.py to view a minimal runnable demo.

Key information and usage requirements for GoSkill

• Project positioning: A standalone, single-process, lightweight goal-driven execution helper, not a distributed scheduling system.
• Applicable boundaries: Suitable for long tasks, complex tasks, and scenarios requiring clear acceptance criteria; not suitable for single-turn Q&A or extremely small synchronous functions.
• Dependency environment: Python environment, with development, testing, and build processes managed via a Makefile.
• License: Supports the Apache-2.0 open-source license, allowing free commercial use and secondary development.
• Current status: An experimental execution-mode prototype, a lightweight wrapper for OpenClaw / Agent workflow.

Core advantages of GoSkill

• Goal-driven: Upgrades a task from "execute once" to "continuously advance around the goal," ensuring the result is acceptable.
• Transparent process: Built-in state tracking mechanism, allowing you to view the number of attempts, duration, and criteria progress at any time.
• Lightweight and non-intrusive: Integrate solely through decorators or class encapsulation, without modifying the original business logic.
• Expectation Alignment: Clarify project boundaries, do not promise enterprise-grade capabilities such as distributed scheduling, and establish credible expectations.

GoSkill's project address

• GitHub repository：https://github.com/AIPMAndy/goskill

Comparison of GoSkill with similar competing products

GoSkill's application scenarios

• Code Engineering Migration: Suitable for large-scale project refactoring or cross-platform migration, such as migrating an Android project to the HarmonyOS system, continuously iterating until it meets hard metrics such as zero compilation errors and test pass rates.
• Batch Data Processing: Suitable for long-running analysis tasks, such as batch analysis of 1,000 financial reports or processing large-scale datasets, using coverage and completeness criteria to advance automatically and avoiding manual round-by-round checks.
• Automated Acceptance Process: Oriented toward CI/CD or testing stages that require clear acceptance criteria, encapsulating "execution—verification—retry" as a closed loop to ensure every deployment meets the preset quality threshold.
• Scientific Research and Iterative Experiments: By defining phased success criteria, automatically track experimental progress and reduce repetitive manual operations.
• Agent Workflow Enhancement: As a long-task execution encapsulation layer for OpenClaw or AI Agents, it compensates for the drawback of a single Skill call "stopping once done", improving the reliability of completing complex tasks.