Offline Script Factory

English | 简体中文

Offline Script Factory is a lightweight skill project that helps a coding agent turn a solved task into a reusable local script bundle.

Instead of asking the agent to solve the same task again and again, the workflow is:

1. Solve a task once.
2. Summarize the stable inputs, outputs, and constraints.
3. Generate an offline script bundle.
4. Run the generated script immediately.
5. If execution fails, fix it and run again.

The project is intentionally small. It does not try to generate a full production CLI framework. Its goal is to help users quickly preserve one-off solutions as local tools they can run later without repeating the same agent conversation.

Quick Start

Install the skill:

git clone <your-repo-url>
cd scripts_factory
.\skills\offline-script-factory\scripts\install.ps1

git clone <your-repo-url>
cd scripts_factory
bash ./skills/offline-script-factory/scripts/install.sh

By default, the installer uses the first available location below:

1. OFFLINE_SCRIPT_FACTORY_SKILLS_DIR
2. CODEX_HOME/skills
3. ~/.codex/skills

You can also install to a custom skills directory explicitly:

.\skills\offline-script-factory\scripts\install.ps1 -SkillsDir "D:\my-skills"

bash ./skills/offline-script-factory/scripts/install.sh --skills-dir "$HOME/.my-agent/skills"

Then invoke it with a prompt such as:

Use $offline-script-factory to turn this completed task into a reusable offline script bundle.

What This Project Includes

• A reusable skill at skills/offline-script-factory/SKILL.md
• A bundle scaffolder at skills/offline-script-factory/scripts/init_offline_bundle.py
• A bundle index generator at skills/offline-script-factory/scripts/update_bundle_index.py
• A metadata validator at skills/offline-script-factory/scripts/validate_bundle_metadata.py
• A checklist for offline feasibility and verification at skills/offline-script-factory/references/offline-automation-checklist.md

Core Principles

• Offline-first: package the local part of the workflow whenever possible.
• Script over repeated prompting: prefer a reusable local artifact over re-solving the same task.
• Built-in help: generated scripts should expose usage through --help or -Help.
• Self-verification: the agent must run the generated script before delivery.
• Repair loop: if the generated script fails, the agent should patch it and rerun it.
• Agent-neutral methodology: the core workflow is written to be usable by different coding agents.

Current Bundle Shape

Unless the task clearly needs more, the generated output should stay minimal:

<bundle-name>/
  <bundle-name>.py | <bundle-name>.ps1
  bundle.spec.json
  config.example.json

bundle.spec.json is the bundle's machine-readable metadata source.

config.example.json is optional and should only be created when stable configuration is useful.

Repository Structure

.
├── skills/
│   └── offline-script-factory/
│       ├── SKILL.md
│       ├── agents/
│       │   └── openai.yaml
│       ├── references/
│       │   └── offline-automation-checklist.md
│       └── scripts/
│           ├── init_offline_bundle.py
│           ├── install.ps1
│           ├── install.sh
│           ├── update_bundle_index.py
│           └── validate_bundle_metadata.py
├── README.md
├── README_CN.md
└── LICENSE

Usage

This repository now includes install scripts for quickly copying the skill into a skills directory.

If you keep multiple generated bundles in one directory, run the index generator to produce a local registry:

python .\skills\offline-script-factory\scripts\update_bundle_index.py --root .\my-bundles

This writes bundles.index.json, which can help future agent sessions inspect existing local tools before generating new ones.

You can also validate bundle metadata before release or reuse:

python .\skills\offline-script-factory\scripts\validate_bundle_metadata.py .\my-bundles

If your agent platform does not support explicit skill invocation, you can still reuse the workflow by reading SKILL.md directly.

Verification Standard

Before a generated script is considered done, the agent should:

1. Run the help command.
2. Run a safe verification command such as --self-test, -SelfTest, or --dry-run.
3. Run a small realistic example when safe.
4. Fix failures and rerun until the script passes or an external blocker is confirmed.

Testing

Run the repository test suite with:

python -m unittest discover -s .\tests -v

Current automated coverage focuses on:

• bundle generation for Python
• bundle generation for PowerShell
• metadata index generation
• metadata validation
• PowerShell installer behavior

Scope

This project is a good fit for:

• file processing
• local automation
• repeatable desktop workflows
• report generation
• small utility scripts

This project is not meant to hide unavoidable online dependencies. If a task still needs a website, cloud API, or hosted model, the script should only package the offline portion and state the remaining online boundary clearly.

Status

This project is currently an early-stage skill prototype focused on workflow quality and repeatability.

License

This repository is released under the MIT License. See LICENSE.