agent-workflow-kit
Agent Workflow Kit — Evaluation-first rules, templates, and Skill packages for AI-assisted software projects. Score risk, choose the smallest safe workflow, and generate practical AGENTS.md guidance for any coding agent.
A bilingual, tool-neutral guide and template kit for scoring a repo's risk (0-16) and picking the right AGENTS.md workflow level for AI coding agents.
Agent Workflow Kit is a documentation-first project for teams that are bringing AI coding agents into real software repositories. Its main output is not code that runs, but a set of guides, templates, and reusable rule files that help a maintainer decide whether their project even needs rules for AI agents, and if so, how strict those rules should be. It is written to be tool-neutral, so it can be used with any large language model or coding agent, not only with one specific product.
The central idea in the README is to evaluate before you change. The project gives you a 0 to 16 risk score table. You score the repository, look up the matching workflow level, and copy a corresponding AGENTS.md template into the repo. There are four levels: Level 0 means no formal AI workflow, suitable when AI is not editing the code or the project is disposable. Levels 1, 2, and 3 add progressively more rules: basic agent rules, then spec guidance and external-action safety, and finally review gates for production or security-sensitive work.
The README frames the kit around five questions a maintainer should answer: does the project need an AI workflow at all, how heavy should it be, when should a spec layer be added, what should the agent verify before claiming work is done, and which external actions need explicit approval. A small table maps common problems, such as agents jumping into code too quickly or work being claimed done without evidence, to the parts of the kit that address them.
The repository is bilingual, with the same guide in English and Chinese. Inside a skills folder are two agent-facing packages, one in each language. Each contains a SKILL.md file with the workflow, a yaml metadata file, and reference files with copy-ready AGENTS.md blocks and an optional catalog of engineering rules. The README shows example prompts you can paste into an agent to invoke them.
A small Ruby script in scripts/check-docs.rb runs locally and in GitHub Actions to validate Markdown formatting, YAML syntax, skill folder structure, and links. Licensing is split: the prose uses CC BY 4.0, while the template blocks are released under CC0 so you can paste them into your own repository without attribution.
Where it fits
- Score your repo with the 0-16 table and copy the matching AGENTS.md block
- Add a Skill package so coding agents follow consistent workflow rules
- Run the Ruby docs check in CI to validate Markdown and YAML
- Adopt review gates for security-sensitive or production code