Skip to main content
AMiSi keeps its AI guidance in ordinary Markdown so you can use it with the coding tool and model you prefer. The repository does not require an AI provider key or ship a model runtime.

​
Start with AGENTS.md

Root AGENTS.md is the portable discovery entry point. It directs your tool to the canonical resources under .ai and explains how to preserve worktree changes, select workflows, and verify completed work. A closer AGENTS.md can add subtree-specific instructions. For example, apps/docs/AGENTS.md applies to the documentation site.

​
Canonical .ai resources

These files are committed to Git and contain no customer credentials or generated model output.

​
Runtime-independent execution

If your tool supports agents, skills, or parallel delegation, it can map the .ai contracts to those native capabilities. Give every isolated worker the complete task and the relevant contract files. If your tool does not support those capabilities, ask it to read the same files and follow them sequentially in the current conversation. The required behavior stays the same.
Keep one worktree writer. Parallelize read-only research and review, but serialize implementation changes through one implementer.

​
Safe workflow

  1. Give the AI a focused request or a task based on .ai/tasks/templates.
  2. Ask it to read AGENTS.md and .ai/README.md.
  3. Select the matching workflow or individual skill.
  4. Review the proposed scope before implementation.
  5. Run targeted checks and finish with:
  1. Review the final diff before committing.