creating-prompts

Create a `.prompt.md` file with frontmatter defining the target agent, description, and argument hint, followed by a body with inputs, a numbered workflow, and output expectations. This skill provides a template and example to scaffold the file.

Supported fields include `name` for a display label, `agent` to route the workflow, `description` for discoverability, `model` for model selection, `argument-hint` for input guidance, and `tools` for overriding default tools when needed.

Use a prompt file when you need a reusable task starter for a repeated workflow. Choose a custom agent when you need a standing expert role, a skill for reusable procedural knowledge, or instructions for always-on repository rules.

Weak output usually comes from missing explicit inputs, bundling multiple unrelated jobs into one prompt, or redefining the agent's role instead of defining the task. The included quality checklist identifies these failure modes before shipping.

Yes, the `tools` frontmatter field can override default tools, but it should only be used when the prompt has a concrete need. Adding tool overrides without justification is flagged as a failure mode in the checklist.