Extend zaly
Start with the smallest mechanism that fits the job:
| Need | Mechanism |
|---|---|
| Explain how a project should be worked on | Project instructions |
| Reuse a prompt with arguments or shell context | Custom commands |
| Format command inputs and conditional text | Templates |
| Teach a reusable workflow loaded on demand | Skills |
| Execute custom integration logic | Plugins |
| Distribute resources together | Resource packs |
User resources apply across projects; project resources travel with a repository. See resource discovery before choosing where to put them.
Keep prompts focused. A short instruction or command is often more maintainable than a plugin. Conversely, use code when something needs deterministic behavior rather than hoping the model will consistently implement it.
Trust
Commands can execute shell snippets, and plugins execute code. Read resources before enabling them. Do not assume the agent's tool permissions sandbox extension code or template evaluation.